Show all abstracts
View Session
- Front Matter: Volume 9301
- Image Processing and Pattern Recognition
Front Matter: Volume 9301
Front Matter: Volume 9301
Show abstract
This PDF file contains the front matter associated with SPIE Proceedings Volume 9301, including the Title Page, Copyright information, Table of Contents, Authors, Introduction (if any), and Conference Committee listing.
Image Processing and Pattern Recognition
Wide-band gas leak imaging detection system using UFPA
Show abstract
The leakage of toxic or hazardous gases not only pollutes the environment, but also threatens people's lives and property safety. Many countries attach great importance to the rapid and effective gas leak detection technology and instrument development. However, the gas leak imaging detection systems currently existing are generally limited to a narrow-band in Medium Wavelength Infrared (MWIR) or Long Wavelength Infrared (LWIR) cooled focal plane imaging, which is difficult to detect the common kinds of the leaking gases. Besides the costly cooled focal plane array is utilized, the application promotion is severely limited. To address this issue, a wide-band gas leak IR imaging detection system using Uncooled Focal Plane Array (UFPA) detector is proposed, which is composed of wide-band IR optical lens, sub-band filters and switching device, wide-band UFPA detector, video processing and system control circuit. A wide-band (3µm~12µm) UFPA detector is obtained by replacing the protection window and optimizing the structural parameters of the detector. A large relative aperture (F#=0.75) wide-band (3μm~12μm) multispectral IR lens is developed by using the focus compensation method, which combining the thickness of the narrow-band filters. The gas leak IR image quality and the detection sensitivity are improved by using the IR image Non-Uniformity Correction (NUC) technology and Digital Detail Enhancement (DDE) technology. The wide-band gas leak IR imaging detection system using UFPA detector takes full advantage of the wide-band (MWIR&LWIR) response characteristic of the UFPA detector and the digital image processing technology to provide the resulting gas leak video easy to be observed for the human eyes. Many kinds of gases, which are not visible to the naked eyes, can be sensitively detected and visualized. The designed system has many commendable advantages, such as scanning a wide range simultaneously, locating the leaking source quickly, visualizing the gas plume intuitively and so on. The simulation experiment shows that the gas IR imaging detection has great advantages and widely promotion space compared with the traditional techniques, such as point-contact or line-contactless detection.
The research on image processing technology of the star tracker
Show abstract
As the core of visual sensitivity via imaging, image processing technology, especially for star tracker, is mainly characterized by such items as image exposure, optimal storage, background estimation, feature correction, target extraction, iteration compensation. This paper firstly summarizes the new research on those items at home and abroad, then, according to star tracker’s practical engineering, environment in orbit and lifetime information, shows an architecture about rapid fusion between multiple frame images, which can be used to restrain oversaturation of the effective pixels, which means star tracker can be made more precise, more robust and more stable.
A real-time multi-scale 2D Gaussian filter based on FPGA
Show abstract
Multi-scale 2-D Gaussian filter has been widely used in feature extraction (e.g. SIFT, edge etc.), image segmentation, image enhancement, image noise removing, multi-scale shape description etc. However, their computational complexity remains an issue for real-time image processing systems. Aimed at this problem, we propose a framework of multi-scale 2-D Gaussian filter based on FPGA in this paper. Firstly, a full-hardware architecture based on parallel pipeline was designed to achieve high throughput rate. Secondly, in order to save some multiplier, the 2-D convolution is separated into two 1-D convolutions. Thirdly, a dedicate first in first out memory named as CAFIFO (Column Addressing FIFO) was designed to avoid the error propagating induced by spark on clock. Finally, a shared memory framework was designed to reduce memory costs. As a demonstration, we realized a 3 scales 2-D Gaussian filter on a single ALTERA Cyclone III FPGA chip. Experimental results show that, the proposed framework can computing a Multi-scales 2-D Gaussian filtering within one pixel clock period, is further suitable for real-time image processing. Moreover, the main principle can be popularized to the other operators based on convolution, such as Gabor filter, Sobel operator and so on.
Design and implementation of ATCA-based 100Gbps DP-QPSK optical signal test instrument
Show abstract
In order to achieve the receiving task of 100Gbps Dual Polarization-Quadrature Phase Shift Keying (DP-QPSK) optical signal acquisition instrument, improve acquisition performance of the instrument, this paper has deeply researched DP-QPSK modulation principles, demodulation techniques and the key technologies of optical signal acquisition. The theories of DP-QPSK optical signal transmission are researched. The DP-QPSK optical signal transmission model is deduced. And the clock and data recovery in high-speed data acquisition and offset correction of multi-channel data are researched. By reasonable hardware circuit design and software system construction, the utilization of high performance Advanced Telecom Computing Architecture (ATCA), this paper proposes a 100Gbps DP-QPSK optical signal acquisition instrument which is based on ATCA. The implementations of key modules are presented by comparison and argumentation. According to the modularization idea, the instrument can be divided into eight modules. Each module performs the following functions. (1) DP-QPSK coherent detection demodulation module; (2) deceleration module; (3) FPGA (Field Programmable Gate Array); (4) storage module; (5) data transmission module; (6) clock module; (7) power module; (8) JTAG debugging, configuration module; What is more, this paper has put forward two solutions to test optical signal acquisition instrument performance. The first scenario is based on a standard STM-256 optical signal format and exploits the SignalTap of QuartusII software to monitor the optical signal data. Another scenario is to use a pseudo-random signal series to generate data, acquisition module acquires a certain amount of data signals, and then the signals are transferred to a computer by the Gigabit Ethernet to analyze. Two testing results show that the bit error rate of optical signal acquisition instrument is low. And the instrument fully meets the requirements of signal receiving system. At the same time this design has an important significance in practical applications.
Multi-source remote sensing image fusion method based on sparse representation
Show abstract
To improve the quality of the fused image, we propose a remote sensing image fusion method based on sparse representation. In the method, first, the source images are divided into patches and each patch is represented with sparse coefficients using an overcomplete dictionary. Second, the larger value of sparse coefficients of panchromatic (Pan) image is set to 0. Third, Then the coefficients of panchromatic (Pan) and multispectral (MS) image are combined with the linear weighted averaging fusion rule. Finally, the fused image is reconstructed from the combined sparse coefficients and the dictionary. The proposed method is compared with intensity-hue-saturation (IHS), Brovey transform (Brovey), discrete wavelet transform (DWT), principal component analysis (PCA) and fast discrete curvelet transform (FDCT) methods on several pairs of multifocus images. The experimental results demonstrate that the proposed approach performs better in both subjective and objective qualities.
Pixel arrangement design of retina-like sensor based on forward motion imaging visual task
Show abstract
Retina-like sensor is a kind of anthropomorphic visual sensor, which mimic the distribution of photoreceptors in the human retina. They are applied in fields of machine vision and target tracking. However, there are few reports on retina-like sensor used for forward-motion imaging. During forward-motion imaging, as the objects being imaged move along the optical axis direction during the integration time, image quality becomes worse towards the border of the image. In order to get clearer image, retina-like sensor are trying to be designed based on the feature of forward-motion imaging. In this paper, firstly, the degraded law of rectilinear sensor used for forward-motion imaging is analyzed, the retina-like sensor model based on the feature of forward-motion imaging are proposed. Secondly, the output image of retina-like sensor and rectilinear sensor used during the forward-motion imaging for different scenes at different degeneration degrees are simulated, respectively. Thirdly, the simulated images of both two sensors are assessed by four different image quality assessment methods including visual information fidelity (VIF), complex wavelet structural similarity index (CW-SSIM), Gabor filtered image contrast similarity (GFCS) and peak signal to noise ratio (PSNR), besides, the data amount of two sensors are compared. Four image quality assessments all demonstrate that image quality of retina-like sensor based on the feature of forward motion imaging is superior to that of rectilinear sensor.
3D Gaussian bin based sparse representation for action recognition
Show abstract
We propose a new representation 3DGBOJ to quickly and precisely classify human action from a series of depth maps. We use Shotton et al's method to predict the best candidate of 3D skeletal joint locations from Kinect depth map. By normalizing and retargeting the human skeleton to a common skeleton, we eliminate the noisy introduced by human agent diversity and view dependent. Some impossible motions are deleted with regard to Kinematics constraint. We design a 3D Gaussian space to map each joint to a bin based sparse feature vector. To weaken the timescale variation, which occurs during the performance with different speed and style, we remove the consecutive repeated vectors. We cluster the motion feature vectors with Affinity Propagation and treat each motion exemplar as a vocabulary in bag of feature (BOF). To better handle overlapping features and contextual dependencies, we trained them over a linear CRFs model. The experiment result shows that our representation maintains appropriate adaptability to variations of different subjects of different gender and size, and with different speed and style from different view.
A new template matching method based on contour information
Show abstract
Template matching is a significant approach in machine vision due to its effectiveness and robustness. However, most of the template matching methods are so time consuming that they can’t be used to many real time applications. The closed contour matching method is a popular kind of template matching methods. This paper presents a new closed contour template matching method which is suitable for two dimensional objects. Coarse-to-fine searching strategy is used to improve the matching efficiency and a partial computation elimination scheme is proposed to further speed up the searching process. The method consists of offline model construction and online matching. In the process of model construction, triples and distance image are obtained from the template image. A certain number of triples which are composed by three points are created from the contour information that is extracted from the template image. The rule to select the three points is that the template contour is divided equally into three parts by these points. The distance image is obtained here by distance transform. Each point on the distance image represents the nearest distance between current point and the points on the template contour. During the process of matching, triples of the searching image are created with the same rule as the triples of the model. Through the similarity that is invariant to rotation, translation and scaling between triangles, the triples corresponding to the triples of the model are found. Then we can obtain the initial RST (rotation, translation and scaling) parameters mapping the searching contour to the template contour. In order to speed up the searching process, the points on the searching contour are sampled to reduce the number of the triples. To verify the RST parameters, the searching contour is projected into the distance image, and the mean distance can be computed rapidly by simple operations of addition and multiplication. In the fine searching process, the initial RST parameters are discrete to obtain the final accurate pose of the object. Experimental results show that the proposed method is reasonable and efficient, and can be used in many real time applications.
Modifications in SIFT-based 3D reconstruction from image sequence
Show abstract
In this paper, we aim to reconstruct 3D points of the scene from related images. Scale Invariant Feature Transform( SIFT) as a feature extraction and matching algorithm has been proposed and improved for years and has been widely used in image alignment and stitching, image recognition and 3D reconstruction. Because of the robustness and reliability of the SIFT’s feature extracting and matching algorithm, we use it to find correspondences between images. Hence, we describe a SIFT-based method to reconstruct 3D sparse points from ordered images. In the process of matching, we make a modification in the process of finding the correct correspondences, and obtain a satisfying matching result. By rejecting the “questioned” points before initial matching could make the final matching more reliable. Given SIFT’s attribute of being invariant to the image scale, rotation, and variable changes in environment, we propose a way to delete the multiple reconstructed points occurred in sequential reconstruction procedure, which improves the accuracy of the reconstruction. By removing the duplicated points, we avoid the possible collapsed situation caused by the inexactly initialization or the error accumulation. The limitation of some cases that all reprojected points are visible at all times also does not exist in our situation. “The small precision” could make a big change when the number of images increases. The paper shows the contrast between the modified algorithm and not. Moreover, we present an approach to evaluate the reconstruction by comparing the reconstructed angle and length ratio with actual value by using a calibration target in the scene. The proposed evaluation method is easy to be carried out and with a great applicable value. Even without the Internet image datasets, we could evaluate our own results. In this paper, the whole algorithm has been tested on several image sequences both on the internet and in our shots.
A moving target detection algorithm based on GMM and improved Otsu method
Show abstract
Based on Gaussian mixture model, an improved detection algorithm, which aimed at updating the real-time character and accuracy of the moving target detection in intelligent video surveillance systems effectively, is elaborated in this paper. It combines the advantages of GMM and improved maximum between class variance method. The algorithm not only improves the speed of detecting targets in the intelligent systems, but also solves the inherent problems efficiently in poor real-time performance and error detection problem. In conclusion, the experiment results demonstrated that the algorithm has an excellent adaptability and anti-interference performance to fit the complicated situation and changing environment.
The FPGA realization of a real-time Bayer image restoration algorithm with better performance
Show abstract
Along with the wide usage of realizing Bayer color interpolation algorithm through FPGA, better performance, real-time processing, and less resource consumption have become the pursuits for the users. In order to realize the function of high speed and high quality processing of the Bayer image restoration with less resource consumption, the color reconstruction is designed and optimized from the interpolation algorithm and the FPGA realization in this article. Then the hardware realization is finished with FPGA development platform, and the function of real-time and high-fidelity image processing with less resource consumption is realized in the embedded image acquisition systems.
An image similarity measure method based on mathematical morphology
Show abstract
An image similarity measure method is proposed. The similarity measure method based on mathematical morphology, ignores the effect of slight distortion or noise on the image similarity, and retains the influence of distortion or loss of the image similarity. The experimental results show that, while the matching image similarity is not reduced, this similarity measure can reduce the matching of image similarity, increasing decision distance. And it helps to improve the performance of image matching, recognition method.
A new correlation parameter extraction method for searching mode sea clutter restraint
Show abstract
Correlation characteristics are important for sea clutter restraint in radar image processing, which is universal in certain features for both radar and optical image. The spatial-temporal dispersion relation contained in sea clutter, which usually fits into the sum of wind friction linear item and gravity wave item, is proved effective for sea clutter restraint recently. A sea clutter restraint method for mobile searching mode observation point, which is prevalent in maritime air-borne platform, is developed from a former restraint method for shore-based stationary radar. A new extraction method for intrinsic dispersion linear item parameter of sea clutter aiming at range profile of searching mode radar system is proposed, which forms the core of the restraint method. The statistical model is founded on spatial-temporal dispersion relation and radar illumination geometry. Simulation results shows Doppler shift of measured wind linear item have a bias from the sum of sea wind and radar platform velocity. A systematic study suggests the bias could be attributed to interaction between wind friction and gravity wave item, and a rather good fitting result is obtained. CFAR detection processing for a set of experimental clutter data shows that this intrinsic dispersion extraction method formula is found to be effective in detection probability enhancement. The restraint method based on the proposed dispersion extraction method could be utilized in mobile maritime surveillance equipments.
Review of development of laser active imaging technology in China and foreign countries
Show abstract
Range-gated laser active imaging technology is an effective way to image detection and precise tracking of remote, dark, and small targets that overcomes the shortcomings of passive visible or infrared imaging technology, thus has important practical value and broad application prospects in the military. The paper based on the analysis of its principle, technical advantages and key technologies focus on the typical systems under atmospheric conditions at home and abroad and the latest research results, and discusses the development trends of this technology.
Improvement of wavelet threshold filtered back-projection image reconstruction algorithm
Show abstract
Image reconstruction technique has been applied into many fields including some medical imaging, such as X ray computer tomography (X-CT), positron emission tomography (PET) and nuclear magnetic resonance imaging (MRI) etc, but the reconstructed effects are still not satisfied because original projection data are inevitably polluted by noises in process of image reconstruction. Although some traditional filters e.g., Shepp-Logan (SL) and Ram-Lak (RL) filter have the ability to filter some noises, Gibbs oscillation phenomenon are generated and artifacts leaded by back-projection are not greatly improved. Wavelet threshold denoising can overcome the noises interference to image reconstruction. Since some inherent defects exist in the traditional soft and hard threshold functions, an improved wavelet threshold function combined with filtered back-projection (FBP) algorithm was proposed in this paper. Four different reconstruction algorithms were compared in simulated experiments. Experimental results demonstrated that this improved algorithm greatly eliminated the shortcomings of un-continuity and large distortion of traditional threshold functions and the Gibbs oscillation. Finally, the availability of this improved algorithm was verified from the comparison of two evaluation criterions, i.e. mean square error (MSE), peak signal to noise ratio (PSNR) among four different algorithms, and the optimum dual threshold values of improved wavelet threshold function was gotten.
A new approach to blind PSF estimation based on Mallows' statistics Cp
Show abstract
We propose an unbiased estimator of the weighted mean squared error — Mallows’ statistics Cp — as a novel criterion for estimating a point spread function (PSF) from the degraded image only. The PSF is obtained by minimizing this new objective functional over a family of Wiener processings. Based on this estimated PSF, we then perform non-blind deconvolution using the popular BM3D algorithm. The Cp-based framework is exemplified with a number of parametric PSF’s, involving a scaling factor that controls the blur size. A typical example of such parametrization is the Gaussian kernel.
The experimental results demonstrate that the Cp-minimization yields highly accurate estimates of the PSF parameters, which also result in a negligible loss of visual quality, compared to that obtained with the exact PSF. The highly competitive results outline the great potential of developing more powerful blind deconvolution algorithms based on the Cp-estimator.
Pose estimation for one-dimensional object with general motion
Show abstract
Our primary interest is in real-time one-dimensional object’s pose estimation. In this paper, a method to estimate general motion one-dimensional object’s pose, that is, the position and attitude parameters, using a single camera is proposed. Centroid-movement is necessarily continuous and orderly in temporal space, which means it follows at least approximately certain motion law in a short period of time. Therefore, the centroid trajectory in camera frame can be described as a combination of temporal polynomials. Two endpoints on one-dimensional object, A and B, at each time are projected on the corresponding image plane. With the relationship between A, B and centroid C, we can obtain a linear equation system related to the temporal polynomials’ coefficients, in which the camera has been calibrated and the image coordinates of A and B are known. Then in the cases that object moves continuous in natural temporal space within the view of a stationary camera, the position of endpoints on the one-dimensional object can be located and also the attitude can be estimated using two end points. Moreover the position of any other point aligned on one-dimensional object can also be solved. Scene information is not needed in the proposed method. If the distance between the endpoints is not known, a scale factor between the object’s real positions and the estimated results will exist. In order to improve the algorithm’s performance from accuracy and robustness, we derive a pain of linear and optimal algorithms. Simulations’ and experiments’ results show that the method is valid and robust with respect to various Gaussian noise levels. The paper’s work contributes to making self-calibration algorithms using one-dimensional objects applicable to practice. Furthermore, the method can also be used to estimate the pose and shape parameters of parallelogram, prism or cylinder objects.
Novel image detail enhancement technology for high dynamic range infrared detector
Show abstract
In this paper, we propose a novel image detail enhancement technology which is well solved the problem of how to suppress the noise and enhance the detail at the same time of the infrared image. This technology is based on the layer separation idea. In nowadays, this idea is studied by many researchers, and many detail enhancement algorithms have been come up through this idea such as the bilateral filter for detail enhancement. According to our research, these algorithms although have the advantages of enhancing the detail without enhancing the noise, they also have the disadvantages of massive calculation, low speed and the worst is the gradient flipping effect which cause the enhanced image distorted. Our solution is based on the Guided Image Filter (GIF) to deal the separated detail layer of an image. The gradient flipping effect will be greatly suppressed with the priority that the GIF is a linear filter. Which means that the processed image will become much closer to the original image. We determine an adaptive weighting coefficient as the filter kernel. After that, we compress the base component into the display range by our modified histogram projection and enhance the detail component using the gain mask of the filter weighting coefficient. At last, we recombine the two parts and quantize the result to 8-bit domain. Experimental verification and detailed realization have been provided in this paper. We also have done significant comparison between our method and the proposed algorithm to show the superiority of our algorithm.
Robust fiber clustering of cerebral fiber bundles in white matter
Show abstract
Diffusion tensor imaging fiber tracking (DTI-FT) has been widely accepted in the diagnosis and treatment of brain diseases. During the rendering pipeline of specific fiber tracts, the image noise and low resolution of DTI would lead to false propagations. In this paper, we propose a robust fiber clustering (FC) approach to diminish false fibers from one fiber tract. Our algorithm consists of three steps. Firstly, the optimized fiber assignment continuous tracking (FACT) is implemented to reconstruct one fiber tract; and then each curved fiber in the fiber tract is mapped to a point by kernel principal component analysis (KPCA); finally, the point clouds of fiber tract are clustered by hierarchical clustering which could distinguish false fibers from true fibers in one tract. In our experiment, the corticospinal tract (CST) in one case of human data in vivo was used to validate our method. Our method showed reliable capability in decreasing the false fibers in one tract. In conclusion, our method could effectively optimize the visualization of fiber bundles and would help a lot in the field of fiber evaluation.
SAR target feature extraction and recognition based multilinear principal component analysis
Show abstract
In this paper, a multilinear principal component analysis (MPCA) algorithm is applied to dimensionality reduction in synthetic aperture radar (SAR) images target feature extraction. Firstly, the MPCA algorithm is used to find the projection matrices in each mode and perform dimensionality reduction in all tensor modes. And then the distances of the feature tensors of the testing and training are computed for classification. Experimental results based on the moving and stationary target recognition (MSTAR) data indicate that compared with the existing methods, such as principal component analysis (PCA), 2-dimensional PCA (2DPCA), and generalized low rank approximations of matrices (GLRAM), the MPCA algorithm achieves the best recognition performance with acceptable feature dimensionality.
Focal length precise measurement method for optics system based on lunar imaging
Show abstract
In this paper an accurate measurement method for optics system based on the lunar imaging is presented, and this method has the following steps. Firstly, the optical imaging system observes the lunar and acquires the image on the ground or in orbit, and records the position and the time simultaneously, with which the distance to the lunar can be computed. Secondly, the initial region of the lunar in the acquired image is decided by the gray value threshold, and the Canny edge detection method with parabola fitting is used to acquire the sub-pixel image edge points. Thirdly, the extracted edge points are used to preliminary fit the lunar disc, and the lunar ring is formed based on the fitted lunar disc expanded two pixels, then the initial coarse fitting disc is acquired according to the maximum number of edge points located in the lunar ring. Fourthly, the sub-pixel lunar disk can be obtained via the least squares fitting on the base of the initial coarse fitting disc. At last, the focal length of the optical imaging system can be computed with the position relationship between the optical imaging system and the lunar. Experiments show that this method has the ability to focal length measurement with high accuracy and frequency. By the means of imaging to the lunar, taking advantage of the long distance, sub-pixel edge detection and fitting for the lunar disc diameter, etc, whether in the full lunar and the waning lunar，the focal length could be measured accurately. It has a wide application prospects both in the developing and in orbit operating stage for optical imaging system.
Accurate matching method of multimodal image based on phase congruency and local mutual information
Show abstract
Multimodal image matching is difficult because of the contrast and intensity difference of the images. For solving this problem, a new matching method based on the phase congruency, histograms of oriented gradients and local normalized mutual information was proposed. The proposed method first extract feature points and edge map based on monogenic signal and phase congruency method which is insensitivity to variations in illumination and contrast. Then, the Monogenic Phase Congruency Edge Descriptor based on edge orientation histogram was generated by gathering the edge information of orientations from edge map around the feature point. For increasing the matched point-pair number, the Multi Candidate Point Matching Method by selecting multi better candidate matching points for each feature point was presented. Finally, the location accuracy was refined using local normalized mutual information method. The experimental results indicated that the proposed method could achieve higher performance in heterogonous image matching, the average matching correct rate up to 88%, is about 3 times of SURF matching method.
Overview and development about image scrambling degree evaluation
Congli Wang,
Zhi-Bin Chen,
Na-Na Ding
Show abstract
With the development of computer and network communication technology, researchers have advanced many image scrambling algorithms to resolve image information security. But how to evaluate the performance of these algorithms has not been studied scientifically. Most of the evaluation ways depend on the original image with large computing work. So it is an important problem to evaluate the image scrambling degree objectively. This paper summarizes the common evaluating parameters, describes their shortcomings and points out the directions about how to evaluate the scrambling degree scientifically further.
An improved infrared image enhancement algorithm based on multi-scale decomposition
Show abstract
Due to the restriction of infrared imaging component and the radiation of atmosphere, infrared images are discontented with image contrast, blurry, large yawp. Aimed on these problems, a multi-scale image enhancement algorithm is proposed. The main principle is as follows: firstly, On the basis of the multi-scale image decomposition, We use an edge-preserving spatial filter that instead of the Gaussion filter proposed in the original version, adjust the scale-dependent factor With a weighted information. Secondly, contrast is equalized by applying nonlinear amplification. Thirdly, subband image is the weighted sum of sampled subband image and subsampled then upsampled subband image by a factor of two. Finally, Image reconstruction was applied. Experiment results show that the proposed method can enhance the original infrared image effectively and improve the contrast, moreover, it also can reserve the details and edges of the image well.
Design of video interface conversion system based on FPGA
Show abstract
This paper presents a FPGA based video interface conversion system that enables the inter-conversion between digital and analog video. Cyclone IV series EP4CE22F17C chip from Altera Corporation is used as the main video processing chip, and single-chip is used as the information interaction control unit between FPGA and PC. The system is able to encode/decode messages from the PC. Technologies including video decoding/encoding circuits, bus communication protocol, data stream de-interleaving and de-interlacing, color space conversion and the Camera Link timing generator module of FPGA are introduced. The system converts Composite Video Broadcast Signal (CVBS) from the CCD camera into Low Voltage Differential Signaling (LVDS), which will be collected by the video processing unit with Camera Link interface. The processed video signals will then be inputted to system output board and displayed on the monitor.The current experiment shows that it can achieve high-quality video conversion with minimum board size.
Error analysis and compensation research of scale factor for MEMS gyroscope
Show abstract
In dynamic condition, scale factor has been one of the main errors for MEMS (micro electromechanical system) gyroscopes. This paper, based on one kind of gyroscope in the airborne optoelectronic pod, studies the variation law of the scale factor and its compensation under different environment temperature and operating speed, and then puts forward to the method of combination of ambient temperature and actual angular velocity when compensating the MEMS gyroscope’s scale factor error. Test result demonstrates that the scale factor error can be effectively suppressed, and compared with compensation method only based on temperature or angular velocity separately, this new method is easy practical and presents better performance.
Fundamental matrix estimation for binocular vision measuring system used in wild field
Show abstract
A method has been proposed to estimate the fundamental matrix of a positing and monitoring binocular vision system with a long working distance and a large field of view. Because of the long working distance and large field of view, images grabbed by this system are seriously blurred, leading to a lack of local features. The edge points are acquired using the Canny algorithm firstly, then the pre-matched points are obtained by the GMM-based points sets registration algorithm, and eventually the fundamental matrix are estimated using the RANSAC algorithm. In actual application, two cameras are 2km away from the object, the fundamental matrix are figured out, and the distance between each point and the corresponding epipolar line is less than 0.8 pixel. Repeated experiments indicate that the average distances between the points and the corresponding epipolar lines are all within 0.3 pixel and the deviations of the distances are all within 0.3 pixel too. This method takes full advantage of the edges in the environment and does not need extra control points, whats more, it can work well in low SNR images.
40-Gbps optical backbone network deep packet inspection based on FPGA
Show abstract
In the era of information, the big data, which contains huge information, brings about some problems, such as high speed transmission, storage and real-time analysis and process. As the important media for data transmission, the Internet is the significant part for big data processing research. With the large-scale usage of the Internet, the data streaming of network is increasing rapidly. The speed level in the main fiber optic communication of the present has reached 40Gbps, even 100Gbps, therefore data on the optical backbone network shows some features of massive data. Generally, data services are provided via IP packets on the optical backbone network, which is constituted with SDH (Synchronous Digital Hierarchy). Hence this method that IP packets are directly mapped into SDH payload is named POS (Packet over SDH) technology. Aiming at the problems of real time process of high speed massive data, this paper designs a process system platform based on ATCA for 40Gbps POS signal data stream recognition and packet content capture, which employs the FPGA as the CPU. This platform offers pre-processing of clustering algorithms, service traffic identification and data mining for the following big data storage and analysis with high efficiency. Also, the operational procedure is proposed in this paper. Four channels of 10Gbps POS signal decomposed by the analysis module, which chooses FPGA as the kernel, are inputted to the flow classification module and the pattern matching component based on TCAM. Based on the properties of the length of payload and net flows, buffer management is added to the platform to keep the key flow information. According to data stream analysis, DPI (deep packet inspection) and flow balance distribute, the signal is transmitted to the backend machine through the giga Ethernet ports on back board. Practice shows that the proposed platform is superior to the traditional applications based on ASIC and NP.
An evolutionary method to achieve stable superpixel tracking
Wenxing Xi,
Xinyi Tang
Show abstract
Object tracking is a hot and hard problem in the computer vision study area.We deal with large objects,which are challenged in many aspects,such as the factors of lighting, size, posture, disturbance, occlusion, and so on.The superpixel tracking method has been proposed to deal with this problem. Unlike many other approaches, it is robust in all the mentioned aspects to some extent. It is very flexible to deal with non-rigid objects just like the meanshift of color histogram does,but can be more advanced, since it takes advantage of the segmented local color histogram. Here we first introduce the adaptive superpixel tracking algorithm, which is comprised by two parts, modeling and confidence mapping using the color features of superpixels.We model them by clustering, just like the "bags of words" method does, and build the cluster confidence.The model is adaptive since it just learns from some latest tracked frames, which can accumulate errors and lead to drift easily. So we propose a refined model, which incorporates the kalman filter's ideas to this problem, by integrating the current model and the new model as an evolutionary one, to better adapt to the object variation and disturbance in subsequent frames, thus achieve more stable tracking. The evolutionary model is achieved by reclustering the cluster centers of the two models, to make new cluster centers and new cluster confidences. We allocate different weight to them, if the current model gets more weight, then the evolutionary model will be more stable, otherwise it will be more adaptive. Finally we give some experiment comparisons between the evolutionary model and the adaptive one. For most cases, when the scene of the object is stable, namely there is no big sudden light change or color change, the evolutionary model outperforms the adaptive one. The reason is that the adaptive one easily learns from other objects. But when the scene suffers big sudden change, the evolutionary model can’t quickly adapt to it and get failed, while the adaptive one may make it. In a word, the method is devised to achieve more stable tracking for stable scene applications.
Research on color constancy computation based on YCbCr color space and gray surface
Show abstract
Color constancy is of important for many computer vision applications, such as image classification, color object recognition, object tracking and so on. But unlike the human visual system, imaging device cannot be able to compute color constant descriptors which do not vary with the color of the illuminant, so solving color constancy problem is necessary. In the calculation of color constancy, illuminant estimation is the key. Because grey surfaces can perfectly reflect the color of the scene illumination, many methods have been proposed to identify grey surfaces to estimate the illuminant. But they either rely on the camera’s parameters, lacking universality, or work inaccurate in worse conditions. In order to solve these problems, in this paper, an iterative method is proposed. The quality of the proposed method is tested and compared to the previous color constancy methods on the Macbeth Chart and two data sets of synthetic and real images. Through MATLAB simulation, experimental pictures and quantitative data for performance evaluation were gotten. The simulated results show that the proposed algorithm is accurate and efficient in identification of the grey surfaces, even in worse condition. And it performs well in color constancy computation on both synthetic and real images.
Detection algorithm for glass bottle mouth defect by continuous wavelet transform based on machine vision
Show abstract
An efficient algorithm based on continuous wavelet transform combining with pre-knowledge, which can be used to detect the defect of glass bottle mouth, is proposed. Firstly, under the condition of ball integral light source, a perfect glass bottle mouth image is obtained by Japanese Computar camera through the interface of IEEE-1394b. A single threshold method based on gray level histogram is used to obtain the binary image of the glass bottle mouth. In order to efficiently suppress noise, moving average filter is employed to smooth the histogram of original glass bottle mouth image. And then continuous wavelet transform is done to accurately determine the segmentation threshold. Mathematical morphology operations are used to get normal binary bottle mouth mask. A glass bottle to be detected is moving to the detection zone by conveyor belt. Both bottle mouth image and binary image are obtained by above method. The binary image is multiplied with normal bottle mask and a region of interest is got. Four parameters (number of connected regions, coordinate of centroid position, diameter of inner cycle, and area of annular region) can be computed based on the region of interest. Glass bottle mouth detection rules are designed by above four parameters so as to accurately detect and identify the defect conditions of glass bottle. Finally, the glass bottles of Coca-Cola Company are used to verify the proposed algorithm. The experimental results show that the proposed algorithm can accurately detect the defect conditions of the glass bottles and have 98% detecting accuracy.
Research on the multi-criteria combination in automatic recognition of marking points
Show abstract
Objective: In the field of computer vision, the technology for the automatic recognition of coded pattern plays an important basic role in the camera calibration process of intrinsic and extrinsic parameters, the binocular image matching process and the three-dimensional reconstruction process. Therefore, in the measurement processing, the successive rate for the automatic recognition of coded pattern must be guaranteed. Method: According to analyzing the geometric information of the coded pattern (the mixed type) and basing on the existing recognition method, a new automatic recognition method is proposed, which is the effective method to solve the multi-points recognition in single image by taking the multi-feature information of the coded pattern as the recognition criteria. Result: Both the new recognition method and the old recognition method are used in identifying the one hundred coded pattern which have been actually collected. The experimental result shows that, not only the new recognition method can achieve accurate identification of coded pattern with the recognition accuracy rate of 100%, but also its processing speed is 2.38 times faster than that in the old recognition method. Conclusion: It is obvious that there are many advantages in the new automatic recognition method, including the high effective recognition, the faster executive speed and independent on the auxiliary decoding process information. The new recognition method of multi-criteria combination can provide a strong guarantee for the realization of every aspect in the work of photogrammetry.
The real-time complex cruise scene motion detection system based on DSP
Show abstract
Dynamic target recognition is an important issue in the field of image processing research. It is widely used in photoelectric detection, target tracking, video surveillance areas. Complex cruise scene of target detection, compared to the static background, since the target and background objects together and both are in motion, greatly increases the complexity of moving target detection and recognition. Based on the practical engineering applications, combining an embedded systems and real-time image detection technology, this paper proposes a real-time movement detection method on an embedded system based on the FPGA + DSP system architecture on an embedded system. The DSP digital image processing system takes high speed digital signal processor DSP TMS320C6416T as the main computing components. And we take large capacity FPGA as coprocessor. It is designed and developed a high-performance image processing card. The FPGA is responsible for the data receiving and dispatching, DSP is responsible for data processing. The FPGA collects image data and controls SDRAM according to the digital image sequence. The SDRAM realizes multiport image buffer. DSP reads real-time image through SDRAM and performs scene motion detection algorithm. Then we implement the data reception and data processing parallelization. This system designs and realizes complex cruise scene motion detection for engineering application. The image edge information has the anti-light change and the strong anti-interference ability. First of all, the adjacent frame and current frame image are processed by convolution operation, extract the edge images. Then we compute correlation strength and the value of movement offset. We can complete scene motion parameters estimation by the result, in order to achieve real-time accurate motion detection. We use images in resolution of 768 * 576 and 25Hz frame rate to do the real-time cruise experiment. The results show that the proposed system achieves real-time processing requirements for the engineering applications.
Human recognition based on head-shoulder contour extraction and BP neural network
Show abstract
In practical application scenarios like video surveillance and human-computer interaction, human body movements are uncertain because the human body is a non-rigid object. Based on the fact that the head-shoulder part of human body can be less affected by the movement, and will seldom be obscured by other objects, in human detection and recognition, a head-shoulder model with its stable characteristics can be applied as a detection feature to describe the human body. In order to extract the head-shoulder contour accurately, a head-shoulder model establish method with combination of edge detection and the mean-shift algorithm in image clustering has been proposed in this paper. First, an adaptive method of mixture Gaussian background update has been used to extract targets from the video sequence. Second, edge detection has been used to extract the contour of moving objects, and the mean-shift algorithm has been combined to cluster parts of target’s contour. Third, the head-shoulder model can be established, according to the width and height ratio of human head-shoulder combined with the projection histogram of the binary image, and the eigenvectors of the head-shoulder contour can be acquired. Finally, the relationship between head-shoulder contour eigenvectors and the moving objects will be formed by the training of back-propagation (BP) neural network classifier, and the human head-shoulder model can be clustered for human detection and recognition. Experiments have shown that the method combined with edge detection and mean-shift algorithm proposed in this paper can extract the complete head-shoulder contour, with low calculating complexity and high efficiency.
A robust point set registration algorithm based on information geometry
Show abstract
Point set registration is a key component in many computer vision tasks. This paper proposes a point set registration algorithm based on information geometry. Point sets to be registration are converting to the statistical manifolds by Gaussian mixture model. The component of mixture model represents the dimension of statistical manifold and point set is a point on manifold. Through conversion, point set registration is reformulated as searching the shortest path between two manifold and we can use the em algorithm which defined by information geometry to get the optimization solution. Experimental results show that the proposed algorithm is robust to noise and outliers, and achieved very good accuracy.
A new thin cloud removal algorithm in single airborne image
Show abstract
The application of high-resolution airborne images becomes more and more extensive. However because of the complexity of atmospheric environment, airborne remote sensing imaging process is easily affected by cloud and mist, which results in airborne image blurred or loss of information. So it is a necessary task to remove effects of cloud to get clearer images before the next application such as image registration. This paper proposes a new method of removing thin cloud cover from single airborne image. This method applies scale space transform to get scale space sequence images. Then we use difference between different levels to extract cloud area. Next, we use gray classification which represents cloud effect degree in the highest level of cloud area. Finally, we use the original image filtered by Laplacian to subtract the last step result. Compared with other thin cloud cover removal methods which include the homomorphic filtering method, wavelet transform method and mathematical morphology by visual evaluation and statistical analysis, the method proposed by this paper proves to be the most efficient way in the processing of thin cloud cover of airborne image.
Facial expression recognition based on fused Feature of PCA and LDP
Show abstract
Facial expression recognition is an important part of the study in man-machine interaction. Principal component analysis (PCA) is an extraction method based on statistical features which were extracted from the global grayscale features of the whole image .But the grayscale global features are environmentally sensitive. In order to recognize facial expression accurately, a fused method of principal component analysis and local direction pattern (LDP) is introduced in this paper. First, PCA extracts the global features of the whole grayscale image; LDP extracts the local grayscale texture features of the mouth and eyes region, which contribute most to facial expression recognition, to complement the global grayscale features of PCA. Then we adopt Support Vector Machine (SVM) classifier for expression classification. Experimental results demonstrate that this method can classify different expressions more effectively and get higher recognition rate compared with the traditional method.
An on-line calibration algorithm for external parameters of visual system based on binocular stereo cameras
Show abstract
Stereo vision is the key in the visual measurement, robot vision, and autonomous navigation. Before performing the system of stereo vision, it needs to calibrate the intrinsic parameters for each camera and the external parameters of the system. In engineering, the intrinsic parameters remain unchanged after calibrating cameras, and the positional relationship between the cameras could be changed because of vibration, knocks and pressures in the vicinity of the railway or motor workshops. Especially for large baselines, even minute changes in translation or rotation can affect the epipolar geometry and scene triangulation to such a degree that visual system becomes disabled. A technology including both real-time examination and on-line recalibration for the external parameters of stereo system becomes particularly important. This paper presents an on-line method for checking and recalibrating the positional relationship between stereo cameras. In epipolar geometry, the external parameters of cameras can be obtained by factorization of the fundamental matrix. Thus, it offers a method to calculate the external camera parameters without any special targets. If the intrinsic camera parameters are known, the external parameters of system can be calculated via a number of random matched points. The process is: (i) estimating the fundamental matrix via the feature point correspondences; (ii) computing the essential matrix from the fundamental matrix; (iii) obtaining the external parameters by decomposition of the essential matrix. In the step of computing the fundamental matrix, the traditional methods are sensitive to noise and cannot ensure the estimation accuracy. We consider the feature distribution situation in the actual scene images and introduce a regional weighted normalization algorithm to improve accuracy of the fundamental matrix estimation. In contrast to traditional algorithms, experiments on simulated data prove that the method improves estimation robustness and accuracy of the fundamental matrix. Finally, we take an experiment for computing the relationship of a pair of stereo cameras to demonstrate accurate performance of the algorithm.
Blur kernel estimate in single noisy image deblurring
Show abstract
Restoring blurred images is challenging because both the blur kernel and the sharp image are unknown, which makes this problem severely under constrained. Recently many single image blind deconvolution methods have been proposed, but these state-of-the-art single image deblurring techniques are still sensitive to image noise, and can degrade their performance rapidly especially when the noise level of the input blurred images increases. In this work, we estimate the blur kernel accurately by applying a series of directional low-pass filters in different orientations to the input blurred image, and effectively constructing the Radon transform of the blur kernel from each filtered image. Finally, we use a robust non-blind deconvolution method with outlier handling, which can effectively reduce ringing artifacts, to generate the final results. Our experimental results on both synthetic and real-world examples show that our method achieves comparable quality to existing approaches on blurry noisy-free images, and higher quality outputs than previous approaches on blurry and noisy images.
Multiple moving objects detection with movement evidence region growing in spatial-temporal domain
Show abstract
Detection of the moving targets is a challenging problem in the fields of computer vision especially on complex circumstance. It plays a very important role for the subsequent advanced task such as tracking and behavior understanding are only related to the moving pixels. To well model the moving detection issue, a novel spatial-temporal multi-scale method is proposed to solve the problem of detecting multiple moving objects on complex background in this paper. Moving objects have multi-scale features both in spatial and temporal domain essentially, which means each object has an optimum temporal-spatial detection window. Hence, the problem of detecting moving objects can be transformed into searching optimal spatial-temporal sub-spaces within different scales. A region growing and splitting recursive algorithm in 3D space and an optimal determinant criterion for estimating motion salience and a real time processing architecture are proposed, which can detect multiple objects at different spatial-temporal scales and extract their features on complex background. Experimental results demonstrated that the proposed method is superior to some of the traditional algorithms and works efficiently in detecting multiple moving objects.
Enhancement algorithm of color fog image based on the brightness adjustment of interception function and adaptive scale
Li-jun Yun,
Yin Gao,
Junsheng Shi,
et al.
Show abstract
To deal with the hue, saturation and brightness problem in the traditional MSRCR fog image enhancement algorithms, Enhancement algorithm of color fog image based on the brightness adjustment of interception function and adaptive scale is proposed. First the image is transformed into the RGB color space. Then according to the each channel pixel values, the grayscale range is stretched by S-cosine curve and the local correction function is introduced. It can calculate the scale of the Gaussian kernel, and then proceeds to do the global correction for the estimates of the reflection component, obtains the multi-scale image by the weighted average. At last, being stretched by the interception function, adjusting the brightness and doing the Gamma correction, which is in order to achieve the image enhancement. Through the subjective observation and objective evaluation, this algorithm has better color fidelity than the traditional MSRCR algorithm in treatment effect.
Low-cost high performance adaptive optics real-time controller in free space optical communication system
Show abstract
This paper proposed a low-cost and high performance adaptive optics real-time controller in free space optical communication system. Real-time controller is constructed with a 4-core CPU with Linux operation system patched with Real-Time Application Interface (RTAI) and a frame-grabber, and the whole cost is below $6000. Multi-core parallel processing scheme and SSE instruction optimization for reconstruction process result in about 5 speedup, and overall processing time for this 137-element adaptive optic system can reach below 100 us and with latency about 50 us by utilizing streamlined processing scheme, which meet the requirement of processing at frequency over 1709 Hz. Real-time data storage system designed by circle buffer make this system to store consecutive image frames and provide an approach to analysis the image data and intermediate data such as slope information.
Improved mean shift target tracking approach under the interference of background
Show abstract
Aiming at the earlier stage tracking of strapdown image terminal guidance system, this paper propose a tracking approach which can not only enhance the distinction between targets and background effectively, but also can constrain the interference of target positioning suffered from background information. Among various target tracking approaches, the Mean Shift tracking algorithm is an excellent one due to its efficiency and simplicity, while it can not effective restrain the disturbance from background information. Thus, in this paper, an only target model background-weighted histogram target tracking algorithm, which can restrain the interference from background information, is presented under the Mean Shift framework. If the histogram of target model and target candidate model are both transformed, the probability of remarkable background features in the target model and target candidate model will simultaneously decrease. Thus it is equivalent to a proportional transformation of the weights obtained by the conventional target representation method. Meanwhile, the Mean Shift iteration formula is invariant to the proportional transformation of weights. Therefore, simultaneously transform the histogram of target model and target candidate model is exactly the same as the Mean Shift tracking with traditional target representation. So the proposed algorithm only transforms the histogram of target model and decreases the probability of target model features that are prominent in the background, but do nothing to target candidate model. Experimental results show that the proposed algorithm can not only restrain the disturbance from background information and improve the tracking accuracy, but also not increases the execution time.
Improved dependent component analysis for hyperspectral unmixing with spatial correlations
Show abstract
In highly mixed hyerspectral datasets, dependent component analysis (DECA) has shown its superiority over other traditional geometric based algorithms. This paper proposes a new algorithm that incorporates DECA with the infinite hidden Markov random field (iHMRF) model, which can efficiently exploit spatial dependencies between image pixels and automatically determine the number of classes. Expectation Maximization algorithm is derived to infer the model parameters, including the endmembers, the abundances, the dirichlet distribution parameters of each class and the classification map. Experimental results based on synthetic and real hyperspectral data show the effectiveness of the proposed algorithm.
Fringes' impetus to positional measurement of stars
Show abstract
In the modern astronomical CCD observation, fringes are annoying problems. It is critical to remove fringes in order to provide properly uniform photometry across the field. Usually a fringe map can be constructed by combining frames and taking medians at every pixel from the corresponding frames’ stack. Furthermore, the fringe map should be scaled based on a target frame in order to remove the fringes precisely. Astrometric work is another different measurement from photometry (for astrophysics), fringes’ impetus to positional determination is often overlooked.
When CCD frames are taken with a slow movement of telescope used, it’s hard to construct a fringe map from data themselves. We extracted the fringe map from other CCD frames in telescope’s different pointings, and thus practiced the approach according to Snodgrass and Carry [1] to derive a scale for a target frame.
Furthermore, the positional measurement was studied for the fringes’ impetus. In more detail, the positional measurements for stars were performed by a well-known 2-D Gaussian fit and were compared before and after de-fringing in the presentation. Our results showed that the biggest positional difference from fringes could be as big as 1 pixel for some faint stars. On average, the mean impetuses (standard deviation) were about 0.03 pixels, 0.25 pixels for bright stars, faint stars, respectively.
An improved corner detection algorithm for image sequence
Show abstract
A SUSAN corner detection algorithm for a sequence of images is proposed in this paper, The correlation matching algorithm is treated for the coarse positioning of the detection area, after that, SUSAN corner detection is used to obtain interesting points of the target. The SUSAN corner detection has been improved. For the situation that the points of a small area are often detected as corner points incorrectly, the neighbor direction filter is applied to reduce the rate of mistakes. Experiment results show that the algorithm enhances the anti-noise performance, improve the accuracy of detection.
Fast 3D reconstruction of tool wear based on monocular vision and multi-color structured light illuminator
Show abstract
Fast 3D reconstruction of tool wear from 2D images has great importance to 3D measuring and objective evaluating tool wear condition, determining accurate tool change and insuring machined part's quality. Extracting 3D information of tool wear zone based on monocular multi-color structured light can realize fast recovery of surface topography of tool wear, which overcomes the problems of traditional methods such as solution diversity and slow convergence when using SFS method and stereo match when using 3D reconstruction from multiple images. In this paper, a kind of new multi-color structured light illuminator was put forward. An information mapping model was established among illuminator's structure parameters, surface morphology and color images. The mathematical model to reconstruct 3D morphology based on monocular multi-color structured light was presented. Experimental results show that this method is effective and efficient to reconstruct the surface morphology of tool wear zone.
Low-light level image de-noising algorithm based on PCA
Show abstract
A de-noising method based on PCA (Principal Component Analysis) is proposed to suppress the noise of LLL (Low-Light Level) image. At first, the feasibility of de-noising with the algorithm of PCA is analyzed in detail. Since the image data is correlated in time and space, it is retained as principal component, while the noise is considered to be uncorrelated in both time and space and be removed as minor component. Then some LLL images is used in the experiment to confirm the proposed method. The sampling number of LLL image which can lead to the best de-noising effects is given. Some performance parameters are calculated and the results are analyzed in detail. To compare with the proposed method, some traditional de-noising algorithm are utilized to suppress noise of LLL images. Judging from the results, the proposed method has more significant effects of de-noising than the traditional algorithm. Theoretical analysis and experimental results show that the proposed method is reasonable and efficient.
Asynchronous data fusion of infrared imaging and laser ranging for target tracking
Jingya Hao,
BingJian Wang,
Jia Li
Show abstract
In a multi-sensor system, sensors often provide data at different rates with different communication delays. To the asynchronous fusion for target tracking with infrared and laser sensors, that provide range and bearing information respectively, there mainly exist two problems need to be solved. One is temporal registration of different sensors since infrared sensor has a shorter sample time than laser sensor. The measurements are dealt with least square method, which makes full use of the high frequency characteristic of infrared sensor. The other is the choice of filtering algorithm because of the correlation of coordinate transformation when measurement equation is described in Cartesian coordinate. Converted measurement Kalman filter (CMKF) is a popular solution to the nonlinear estimation problem expect nonlinear filtering algorithm. Simulations of the algorithms adopted in the paper were carried out to track the target and the performance of different algorithms was compared. The results show that the algorithm based on CMKF yields better track performance than the fusion algorithm based on standard Kalman filter (KF) and a little worse than the algorithm based on unscented Kalman filter (UKF).
Multi-sensor image registration based on visual attention
Show abstract
Inspired by the process of manual registration, a method based on visual attention is proposed in this paper for multi-sensor image registration. In the first stage, the corner points are selected from both of the multi-sensor images using multi-scale Harris detector and then the outlines are extracted by Gabor filter. In the second stage, the selected points are described based on the contour images to find the matching pairs. Finally, the parameters of the affine transformation model between the images are obtained according to the matching pairs. Pairs of visible and infrared images are used to evaluate the performance of the proposed algorithm and SIFT algorithm. Experimental results show that the proposed method can achieve good performance for registering visible and infrared images.
Electronic image stabilization algorithm based on PCA-SIFT feature matching and self-adaptive high-pass filtering
Show abstract
As the electronic image stabilization (EIS) algorithm based on SIFT feature matching has the problem of complex computation and time consuming, a modified EIS algorithm based on PCA-SIFT feature matching and self-adaptive high-pass filtering is proposed in this paper. Firstly, feature points are extracted by using PCA-SIFT algorithm in reference frame and current frame. And the corresponding points are matched between these two images. Then the Random Sample Consensus (RANSAC) algorithm is used to eliminate the error matching pairs to reduce the influence of local motion in the scene and improve the estimation accuracy of global motion parameters. Finally, the random dithering parameters obtained by self-adaptive high-pass filtering are used to compensate the current frames. And the size of filter is adjusted automatically according to dithering frequency to prevent the overstabilization or understabilization. Experimental results show that the algorithm proposed in this paper can effectively remove vectors caused by random dithering and obtain a stable video.
SAR Sea ice classification which is based on wavelet character
Show abstract
For sea ice monitoring , automatic classification of SAR image has great signification. Due of coherence, adjacent pixels of the gray would change randomly in rader echo signals,which causes traditional features can’t work well in SAR sea ice classification. The energy of coherent speckle noise is concentrated in the high frequency. The wavelet transform can decompose signal into different components in the frequency domain, which providing an opportunity to analyze the signal locally.In this paper, a mothod of sea ice classification is adopted, which is based on low-frequency sub-band wavelet feature. The result shows this method reduces the noise influence and improves inaccurate classification caused by noise.
Moving target detection method based on polarization characteristics under the condition of moving detector
Show abstract
Moving target detection is a important field of image processing. Because the existing algorithms are vulnerable to background disturbance’s interference in the case of moving detector, a moving target detection algorithm based on target’s polarization characteristics under the condition of the moving detector has been proposed in this paper. Firstly, for the problem of the target detection such as the movement of background and parallax which are caused by the moving detector, a moving target detection method under the condition of moving detector has been proposed. The method gets detector’s motion estimation and compensation parameters by using image feature points matching method, and uses background updating method to achieve target’s detection. Then LK optical flow method is used to get target’s movement information and detector’s movement information and model the target’s and background’s movement information. Eventually this method calculates the relevance of the background’s and target’s movement information model to achieve target detection. Secondly, for the moving target detection method could not solve the problem of Background disturbance which interferes the detection result, a target detection method fused target polarization characteristics has been proposed on the basis of moving target detection method under the conditions of the rotation detector. This method realizes the target detection algorithm based on target’s polarization characteristics under the condition of moving detector, by pre-processing the polarization images to solve the parallax’s effect, clustering and segmenting the pretreated polarization image to extract polarized target, and fusing the moving target detection method. The experiment result shows that this method can effectively detect moving targets with a strong polarization characteristics in the scene, while suppressing the interference brought by strong polarized but still region and weak polarized background disturbance in the sense.
Fast obstacle detection based on multi-sensor information fusion
Show abstract
Obstacle detection is one of the key problems in areas such as driving assistance and mobile robot navigation, which cannot meet the actual demand by using a single sensor. A method is proposed to realize the real-time access to the information of the obstacle in front of the robot and calculating the real size of the obstacle area according to the mechanism of the triangle similarity in process of imaging by fusing datum from a camera and an ultrasonic sensor, which supports the local path planning decision. In the part of image analyzing, the obstacle detection region is limited according to complementary principle. We chose ultrasonic detection range as the region for obstacle detection when the obstacle is relatively near the robot, and the travelling road area in front of the robot is the region for a relatively-long-distance detection. The obstacle detection algorithm is adapted from a powerful background subtraction algorithm ViBe: Visual Background Extractor. We extracted an obstacle free region in front of the robot in the initial frame, this region provided a reference sample set of gray scale value for obstacle detection. Experiments of detecting different obstacles at different distances respectively, give the accuracy of the obstacle detection and the error percentage between the calculated size and the actual size of the detected obstacle. Experimental results show that the detection scheme can effectively detect obstacles in front of the robot and provide size of the obstacle with relatively high dimensional accuracy.
Infrared small target enhancement by using sequential top-hat filters
Jianjun Zhou,
Haoyin Lv,
Fugen Zhou
Show abstract
Generally in the infrared images, the targets have low contrast with the background, which makes the detection of the small targets difficult. To improve the detectability of the infrared small targets, this paper presents a novel algorithm for infrared small target enhancement by using sequential top-hat filters. Moreover, the proposed algorithm has been compared with several existing algorithms. The experimental results indicate that sequential top-hat filters could well enhance the infrared small targets and effectively suppress the background clutters.
Algorithm about speckle noise filtering in active laser imaging
Show abstract
There is much speckle noise in images of active laser imaging which could decrease image quality. Here, a multidirectional line-type structure weighted morpholoical (MLWM) filter is designed to suppress this speckle noise. The experimental results show that the new MLWM filter has the best performance in terms of noise suppression and the capability of preserving detailed information of images in order for the subsequent processing of imaging system.
Investigation on improved infrared image detail enhancement algorithm based on adaptive histogram statistical stretching and gradient filtering
Show abstract
Duo to infrared image with low contrast, big noise and unclear visual effect, target is very difficult to observed and identified. This paper presents an improved infrared image detail enhancement algorithm based on adaptive histogram statistical stretching and gradient filtering (AHSS-GF). Based on the fact that the human eyes are very sensitive to the edges and lines, the author proposed to extract the details and textures by using the gradient filtering. New histogram could be acquired by calculating the sum of original histogram based on fixed window. With the minimum value for cut-off point, author carried on histogram statistical stretching. After the proper weights given to the details and background, the detail-enhanced results could be acquired finally. The results indicate image contrast could be improved and the details and textures could be enhanced effectively as well.
Study on numerical evaluation method of sky background luminance
Show abstract
Since sky background radiation luminance is a critical parameter of atmospheric optics, it is very important for space target detection and identification. In order to study sky background radiation luminance characteristic, the factors that influenced sky background radiation luminance were analyzed. A method which is used to evaluate sky background radiation luminance based on Image Color Index was put forward. It is valid and possible in the initial test. The method can be compared with other measurement method of sky background radiation luminance, It is as the datum which also could be used to analyses and contrast spectrum characteristic of the atmospheric aerosol particle.
A fast particle filter object tracking algorithm by dual features fusion
Shou-wei Zhao,
Wei-ming Wang,
Sa-sa Ma,
et al.
Show abstract
Under the particle filtering framework, a video object tracking method described by dual cues extracting from integral histogram and integral image is proposed. The method takes both the color histogram feature and the Harr-like feature of the target region as the feature representation model, tracking the target region by particle filter. In the premise of ensuring the real-time responsiveness, it overcomes the shortcomings of poor precision, large fluctuations, light sensitive defects and so on by only relying on histogram feature tracking. It shows high efficiency by tracking the target object in multiple video sequences. Finally, it is applied in the augmented reality assisted maintenance prototype system, which proves that the method can be used in the tracking registration process of the augmented reality system based on natural feature.
Study on mean shift tracking algorithm combined with tracking differentiator
Show abstract
The classical Mean shift algorithm for target tracking, when it is used to track window in a complex background, appears too vulnerable to avoid window jitter, which leads to a failure of accurately tracking the target; while in tracking the target of fast-moving video sequences, the loss of tracking targets can not be avoid. As an improvement based on the classical Mean shift algorithm, the proposed algorithm with characteristic of combining tracking differentiator (TD) can eliminate window jitter and predict the target location with the help of TD. Experiments demonstrate that the proposed algorithm is capable of getting over deficiency of the original algorithm and holding improved stability and robustness.
Research on simulation of target echo of laser fuse in the near field
Show abstract
In order to study the near-field target characteristic of the laser fuse, an algorithm based on the relationship of bidirectional reflectance distribution function and laser radar cross section per unit area is proposed to calculate the echo power of laser fuse in the near-field. The main research work in this paper involves the followings (1)Based on the theory of beam division, a mathematical description of the angular distribution of the detonator laser beam is given to depicted the mathematical model of Gaussian beam. (2)By using the scattering characteristics of rough surface as well as the geometry mesh model of the target, the relation formula between received power and transmitted power of remote system for a facet is derived. (3)Establishing the missile-target encounter model though the conversion from different coordinate systems. Then calculate the echo power of laser fuse by integrating those of the geometrical elements which are illuminated by laser beam during missile target encounter. Consequently, the received power in each channels of the laser fuse can be calculated. In addition, the proposed theoretical model in this paper is calibrated by actually-measured data. And the emulation results are with a good agreement with measured results. Based on the theoretical analysis methods proposed in former chapters, we have developed a program to compute the echo power. Finally , we consider a simplified missile model, and compute its echo power under different angle and different material as well as different miss distance and different target miss in azimuth. The results show that scattering peaks correspond to the points of the wings of the missile. In addition, the results change obviously when using different material .For instance, the results with aluminum material are almost 10 times than that of white paint when ignoring the influence of atmospheric attenuation. At the same time, the results are different under the different miss distance as well as target miss in azimuth. Numerical results prove the proposed method high efficiency and preciseness. It would be especially valuable in engineering application.
The color measurement system for spot color printing basing multispectral camera
Show abstract
Color measurement and control of printing has been an important issue in computer vision technology . In the past,
people have used density meter and spectrophotometer to measure the color of printing product. For the color
management of 4 color press, by these kind meters, people can measure the color data from color bar printed at the side
of sheet, then do ink key presetting. This way have wide application in printing field. However, it can not be used in the
case that is to measure the color of spot color printing and printing pattern directly. With the development of
multispectral image acquisition, it makes possible to measure the color of printing pattern in any area of the pattern by
CCD camera than can acquire the whole image of sheet in high resolution. This essay give a way to measure the color of
printing by multispectral camera in the process of printing. A 12 channel spectral camera with high intensity white LED
illumination that have driven by a motor, scans the printing sheet. Then we can get the image, this image can include
color and printing quality information of each pixel, LAB value and CMYK value of each pixel can be got by
reconstructing the reflectance spectra of printing image. By this data processing, we can measure the color of spot color
printing and control it. Through the spot test in the printing plant, the results show this way can get not only the color bar
density value but also ROI color value. By the value, we can do ink key presetting, that makes it true to control the spot
color automatically in high precision.
Research on imaging ranging algorithm base on constraint matching of trinocular vision
Show abstract
Binocular stereo vision is a common passive ranging method, which directly simulates the approach of human visual. It
can flexibly measure the stereo information in a complex condition. However there is a problem that binocular vision
ranging accuracy is not high , one of the reasons is the low precision of the stereo image pairs matching . In this paper,
based on trinocular vision imaging ranging algorithm of constraint matching, we use trinocular visual ranging system
which is composed of three parallel placed cameras to image and achieve distance measurement of the target. we use
calibration method of Zhang to calibrate the cameras, firstly, the three cameras are calibrated respectively, then using the
results to get three groups binocular calibration. Thereby the relative position information of each camera are obtained.
The using of the information obtained by the third camera can reduce ambiguity of corresponding points matching in a
Binocular camera system. limiting search space by the epipolar constraint and improve the matching speed, filtering
the distance information , eliminate interference information which brings by the feature points on the prospect and
background to obtain a more accurate distance result of target. Experimental results show that, this method can overcome
the limitations of binocular vision ranging , effectively improving the range accuracy.
Lane detection and tracking based on improved Hough transform and least-squares method
Show abstract
Lane detection and tracking play important roles in lane departure warning system (LDWS). In order to improve the
real-time performance and obtain better lane detection results, an improved algorithm of lane detection and tracking based
on combination of improved Hough transform and least-squares fitting method is proposed in this paper. In the image
pre-processing stage, firstly a multi-gradient Sobel operator is used to obtain the edge map of road images, secondly
adaptive Otsu algorithm is used to obtain binary image, and in order to meet the precision requirements of single pixel, fast
parallel thinning algorithm is used to get the skeleton map of binary image. And then, lane lines are initially detected by
using polar angle constraint Hough transform, which has narrowed the scope of searching. At last, during the tracking
phase, based on the detection result of the previous image frame, a dynamic region of interest (ROI) is set up, and within
the predicted dynamic ROI, least-squares fitting method is used to fit the lane line, which has greatly reduced the algorithm
calculation. And also a failure judgment module is added in this paper to improve the detection reliability. When the
least-squares fitting method is failed, the polar angle constraint Hough transform is restarted for initial detection, which has
achieved a coordination of Hough transform and least-squares fitting method. The algorithm in this paper takes into
account the robustness of Hough transform and the real-time performance of least-squares fitting method, and sets up a
dynamic ROI for lane detection. Experimental results show that it has a good performance of lane recognition, and the
average time to complete the preprocessing and lane recognition of one road map is less than 25ms, which has proved that
the algorithm has good real-time performance and strong robustness.
Polygon star identification based on ant colony algorithm
Show abstract
In order to enhance the rate of star identification under different view fields and reduce memory storage,
this paper presents a polygon star identification based on ACO algorithm .First, fast cluster analysis. Second,
calculate argument for each guide star, using the advantages of ACO in fast path optimization to complete
building feature polygon. Third, comparing optimization results and optimization data of guide database to
realize match and identifying. Through the simulation shows that the above method can simplify searching
process and structure of storage. It can promise the completeness of characteristic patterns of star image.
The robustness and reliability are better than traditional triangle identification.
Fruit fly optimization based least square support vector regression for blind image restoration
Show abstract
The goal of image restoration is to reconstruct the original scene from a degraded observation. It is a critical and
challenging task in image processing. Classical restorations require explicit knowledge of the point spread function and a
description of the noise as priors. However, it is not practical for many real image processing. The recovery processing
needs to be a blind image restoration scenario. Since blind deconvolution is an ill-posed problem, many blind restoration
methods need to make additional assumptions to construct restrictions. Due to the differences of PSF and noise energy,
blurring images can be quite different. It is difficult to achieve a good balance between proper assumption and high
restoration quality in blind deconvolution. Recently, machine learning techniques have been applied to blind image
restoration. The least square support vector regression (LSSVR) has been proven to offer strong potential in estimating
and forecasting issues. Therefore, this paper proposes a LSSVR-based image restoration method. However, selecting the
optimal parameters for support vector machine is essential to the training result. As a novel meta-heuristic algorithm, the
fruit fly optimization algorithm (FOA) can be used to handle optimization problems, and has the advantages of fast
convergence to the global optimal solution. In the proposed method, the training samples are created from a
neighborhood in the degraded image to the central pixel in the original image. The mapping between the degraded image
and the original image is learned by training LSSVR. The two parameters of LSSVR are optimized though FOA. The
fitness function of FOA is calculated by the restoration error function. With the acquired mapping, the degraded image
can be recovered. Experimental results show the proposed method can obtain satisfactory restoration effect. Compared
with BP neural network regression, SVR method and Lucy-Richardson algorithm, it speeds up the restoration rate and
performs better. Both objective and subjective restoration performances are studied in the comparison experiments.
Facial expression recognition based on image Euclidean distance-supervised neighborhood preserving embedding
Show abstract
High-dimensional data often lie on relatively low-dimensional manifold, while the nonlinear geometry of that
manifold is often embedded in the similarities between the data points. These similar structures are captured by
Neighborhood Preserving Embedding (NPE) effectively. But NPE as an unsupervised method can’t utilize class
information to guide the procedure of nonlinear dimensionality reduction. They ignore the geometrical structure
information of local data points and the spatial information of pixels, which leads to the failure of classification. For this
problem, a feature extraction method based on Image Euclidean Distance-Supervised NPE (IED-SNPE) is proposed, and
is applied to facial expression recognition. Firstly, it employs Image Euclidean Distance (IED) to characterize the
dissimilarity of data points. And then the neighborhood graph of the input data is constructed according to a certain kind
of dissimilarity between data points. Finally, it fuses prior nonlinear facial expression manifold of facial expression
images and class-label information to extract discriminative features for expression recognition. In the classification
experiments on JAFFE facial expression database, IED-SNPE is used for feature extraction and compared with NPE,
SNPE, and IED-NPE. The results reveal that IED-SNPE not only the local structure of expression manifold preserves
well but also explicitly considers the spatial relationships among pixels in the images. So it excels NPE in feature
extraction and is highly competitive with those well-known feature extraction methods.
The multi-spectral radiation influence research between buildings
Show abstract
In order to improve the accuracy of multi-spectral scene simulation, and avoid resource waste of unnecessary
computing, some researches on the radiation influence between buildings during the multi-spectral simulation in the
waveband 3-5μm and 8-12μm have been done, so as to provide theoretical support for whether it is needed to compute
the radiation influence between buildings in the multi-spectral simulation. This paper determines the primary factors that
affect the degree of radiation influence between buildings, determines the effect that the sun direct radiation to the
radiation influence between buildings, derives the computation formula for radiation influence between buildings in a
specific scene from many basic common heat radiation formula and simulates the scene radiation in multi-spectral in the
specific scene. Finally, the importance of radiation influence between buildings comparing to the entire scene simulation
radiation was evaluated based on numerical calculation. The numerical calculation results show that the radiation
influence between buildings in waveband 3-5μm can be ignored when the sun direct radiation exists, which can’t be
ignored in waveband 8-12μm. In the waveband 8-12μm, the radiation influence between nearby buildings is great in
waveband 8μm, 9μm and 10μm, more than 10% comparing to the buildings’ self radiation, which is small in waveband
11μm and 12μm, less than 4%.
A fast panoramic mosaic method based on position sensor
Show abstract
For easily viewing and operating in The IRST system, it is necessary to montage the 360°panoramic image orderly and
correctly. The paper introduces a fast Panoramic mosaic method for the Infrared Search and Track (IRST) system. First of
all, zero position in azimuth is determined from position sensor. Then relative position of every image is obtained by the
position sensor. Next, the accurate position of image is calculated by integral time of the IR camera. Thus, the panoramic
image mosaic are montaged. This method works more quickly and accurately. The innovative point is obtaining accurate
position information of image making use of position sensor and integral time of the IR camera.
An improved multi-scale autoconvolution transform
Show abstract
Affine invariant feature computing method is an important part of statistical pattern recognition due to the
robustness, repeatability, distinguishability and wildly applicability of affine invariant feature. Multi-Scale
Autoconvolution (MSA) is a transformation proposed by Esa Rathu which can get complete affine invariant feature.
Rathu proved that the linear relationship of any four non-colinear points is affine invariant. The transform is based on a
probabilistic interpretation of the image function. The performance of MSA transform is better on image occlusion and
noise, but it is sensitive to illumination variation. Aim at this problem, an improved MSA transform is proposed in this
paper by computing the map of included angle between N-domain vectors. The proposed method is based on the
probabilistic interpretation of N-domain vectors included angle map. N-domain vectors included angle map is built
through computing the vectors included angle where the vectors are composed of the image point and its N-domain
image points. This is due to that the linear relationship of included angles between vectors composed of any four
non-colinear points is an affine invariance. This paper proves the method can be derived in mathematical aspect. The
transform values can be used as descriptors for affine invariant pattern classification. The main contribution of this
paper is applying the N-domain vectors included angle map while taking the N-domain vector included angle as the
probability of the pixel. This computing method adapts the illumination variation better than taking the gray value of
the pixel as the probability. We illustrate the performance of improved MSA transform in various object classification
tasks. As shown by a comparison with the original MSA transform based descriptors and affine invariant moments, the
proposed method appears to be better to cope with illumination variation, image occlusion and image noise.
Hemagglutinin outer contour detection methods based on regular hexagon bar template
Show abstract
In order to extract hemagglutinin outer contour accurately in the hemagglutinin image,
analyzes the hemagglutinin protein content by the size of detected contour, presents a regular hexagon
bar circle detection algorithm which uses regular hexagon bar detection template to detect outer
contour of the hemagglutinin. Firstly, the hemagglutinin image thresholded by using OTSU adaptive
thresholding method; and then using regular hexagon bar detection template method to rough align
hemagglutinin after thresholded, intersection of detection template and the hemagglutinin contour area
is attained, the noise near hemagglutinin contour is reduced by using the standardization relationship of
the hexagon bars, so the hemagglutinin pixels are accurately obtained; finally the hemagglutinin outer
contour information is gained by the geometric relationship of pixels, the hemagglutinin position is
achieved precisely. The experimental results show that: the contour detection error due to the density
uneven and the edge unclearly of hemagglutinin image protein is better reduced, the detection accuracy
is increased by a factor of 0.47, detection speed is increased by a factor of 0.56.The hemagglutinin
contour can be dected stablely, fastly, accurately and the is significant to the study of the hemagglutinin
protein content.
The registration of dual-modality ship target images based on edge extraction
Show abstract
In this paper, we study the problem of visible and IR(infrared) ship target image registration with scale changes. We mainly focus on the infrared and visible image feature extraction and matching method. A method based on Force Field Transformation is used to determine the ship target contour area. Canny edge detection method is applied to obtain the edge features. During the process of image registration, we take the cross-correlation as the similarity measure and propose an improved Powell algorithm based on multi-scale searching to optimize the registration parameters. Through the edge fusion results, we can see the corresponding edges are almost overlapped, indicating that the method could achieve satisfying results. Also the average error distance of match is less than one pixel.
The method for detecting diffusion ring diameter in Hemagglutinin measuring
Show abstract
The diffuser ring diameter measurement is the most critical in hemagglutinin Measuring. The traditional methods,
such as a vernier caliper or high-definition scanned images are subjective and low for the measurement data reliability.
Propose high-resolution diffusion ring image for drop-resolution processing, adaptive Canny operator and local detection
method to extract complete and clear diffusion ring boundaries, and finally make use of polynomial interpolation
algorithm to make diffusion ring outer boundary pixel coordinates achieve sub-pixel accuracy and the least-squares
fitting circle algorithm to calculate the precise center of the circle and the diameter of the diffuser ring. Experimental
results show that the method detection time is only 63.61ms, which is a faster speed; diffuser ring diameter estimation
error can achieve 0.55 pixel, high stability in experimental data. This method is adapted to the various types of influenza
vaccine hemagglutinin content measurements, and has important value in the influenza vaccine quality detection.
People counting in classroom based on video surveillance
Show abstract
Currently, the switches of the lights and other electronic devices in the classroom are mainly relied on manual control, as
a result, many lights are on while no one or only few people in the classroom. It is important to change the current
situation and control the electronic devices intelligently according to the number and the distribution of the students in
the classroom, so as to reduce the considerable waste of electronic resources. This paper studies the problem of people
counting in classroom based on video surveillance. As the camera in the classroom can not get the full shape contour
information of bodies and the clear features information of faces, most of the classical algorithms such as the pedestrian
detection method based on HOG (histograms of oriented gradient) feature and the face detection method based on
machine learning are unable to obtain a satisfied result. A new kind of dual background updating model based on sparse
and low-rank matrix decomposition is proposed in this paper, according to the fact that most of the students in the
classroom are almost in stationary state and there are body movement occasionally. Firstly, combining the frame
difference with the sparse and low-rank matrix decomposition to predict the moving areas, and updating the background
model with different parameters according to the positional relationship between the pixels of current video frame and
the predicted motion regions. Secondly, the regions of moving objects are determined based on the updated background
using the background subtraction method. Finally, some operations including binarization, median filtering and
morphology processing, connected component detection, etc. are performed on the regions acquired by the background
subtraction, in order to induce the effects of the noise and obtain the number of people in the classroom. The experiment
results show the validity of the algorithm of people counting.
Image depth estimation from compressed sensing theory
Show abstract
Depth information of the image is really necessary information to reconstruct a 3-dimensional object.
The classical methods of depth estimation are generally divided into two categories: active and passive
methods. The active methods requires the additional lighting equipment, passive methods also have a
series of problems .They require a plurality of images obtained by capturing a plurality of viewpoints ,
and determine the locating occlusion boundary , etc., and hence the depth estimation has been a
challenging problem in the research field of computer vision.1 Because of the depth information of the
image has a natural sparse features, this paper uses a passive approach, the signal of sparse priori based
on compressed sensing theory is used to estimate the depth of the image, without capturing multiple
images, using a single input image can obtain a high quality depth map. Experimental results show that
the depth map obtaining by our method, compared to the classical passive method, the contour
sharpness, the depth of detail information and the robustness of noise are more advantages. The method
also can be applied to re-focus the defocused images, and automatic scene segmentation and other
issues, ultimately may have broad application prospects in the reconstruction of true 3-dimensional
objects.
Facial expression recognition based on improved DAGSVM
Show abstract
For the cumulative error problem because of randomization sequence of traditional DAGSVM(Directed Acyclic Graph Support Vector Machine) classification, this paper presents an improved DAGSVM expression recognition method. The method uses the distance of class and the standard deviation as the measure of the classer, which minimize the error rate of the upper structure of the classification. At the same time, this paper uses the method which combines discrete cosine transform (Discrete Cosine Transform, DCT) with Local Binary Pattern(Local Binary Pattern，LBP) ,to extract expression feature and be the input to improve the DAGSVM classifier for recognition. Experimental results show that compared with other multi-class support vector machine method, improved DAGSVM classifier can achieve higher recognition rate. And when it’s used at the platform of the intelligent wheelchair, experiments show that the method has a better robustness.
Spatial distorted target recognition based on improved MACH filter
Show abstract
Joint transform correlator (JTC) can make targets recognized and located accurately, but the bottleneck technique of
JTC is how to recognize spatial distorted targets in cluttered scene. This has restricted the development of the pattern
recognition with JTC to a great extent. In order to solve the problem, improved maximum average correlation height
(MACH) filter algorithm is presented in this paper. The MACH algorithm has powerful capability of recognition for
spatial distorted targets (rotation and scale changed etc.). The controlling parameters of the synthesized filter are
optimized in this paper, which makes the filter have higher distortion tolerance and can suppress cluttered noise
effectively. When improved MACH filter algorithm in frequency domain is projected to space domain, the MACH
reference template image can be obtained which includes various forms of distorted target image. Based on amounts of
computer simulation and optical experiments, MACH reference template is proved to have the capability of sharpening
the correlation peaks and expanding recognizing scope for distorted targets in cluttered scene. MATLAB software is
applied to produce MACH reference image for the detected target images and conduct simulation experiments for its
powerful calculation capability of matrix. In order to prove the feasibility of MACH reference in JTC and determine the
recognition scope, experiments for an aircraft target in the sky are carried out. After the original image is processed by
edge extraction, a MACH filter reference template is obtained in space domain from improved MACH filter in frequency
domain. From simulation experiments, the improved MACH filter is proved to have the feasibility of sharpening
correlation peaks for distorted targets. Optical experiments are given to verify the effectiveness further. The experiments
show the angular distortion tolerance can reach up to ±15 degrees and scale distortion tolerance can reach up to ±23%.
Within this scope, the spatial distorted aircraft can be recognized effectively. The actual effect of the improved MACH
filter algorithm has been confirmed very well.
Application of automatic threshold in dynamic target recognition with low contrast
Show abstract
Hybrid photoelectric joint transform correlator can realize automatic real-time recognition with high precision through the combination of optical devices and electronic devices. When recognizing targets with low contrast using photoelectric joint transform correlator, because of the difference of attitude, brightness and grayscale between target and template, only four to five frames of dynamic targets can be recognized without any processing. CCD camera is used to capture the dynamic target images and the capturing speed of CCD is 25 frames per second. Automatic threshold has many advantages like fast processing speed, effectively shielding noise interference, enhancing diffraction energy of useful information and better reserving outline of target and template, so this method plays a very important role in target recognition with optical correlation method. However, the automatic obtained threshold by program can not achieve the best recognition results for dynamic targets. The reason is that outline information is broken to some extent. Optimal threshold is obtained by manual intervention in most cases. Aiming at the characteristics of dynamic targets, the processing program of improved automatic threshold is finished by multiplying OTSU threshold of target and template by scale coefficient of the processed image, and combining with mathematical morphology. The optimal threshold can be achieved automatically by improved automatic threshold processing for dynamic low contrast target images. The recognition rate of dynamic targets is improved through decreased background noise effect and increased correlation information. A series of dynamic tank images with the speed about 70 km/h are adapted as target images. The 1st frame of this series of tanks can correlate only with the 3rd frame without any processing. Through OTSU threshold, the 80th frame can be recognized. By automatic threshold processing of the joint images, this number can be increased to 89 frames. Experimental results show that the improved automatic threshold processing has special application value for the recognition of dynamic target with low contrast.
Study of flow field of burning particles in a pyrotechnic flame based on particle image and particle velocity
Show abstract
Studying the burning particles in the pyrotechnic flame is important to acquire the decomposition mechanism and
spectral radiance of pyrotechnics. The high speed video (HSV) and particle image velocimetry (PIV) were used in
this paper to analyze the flow field and velocity of burning particles in the flame of pyrotechnics. The binary image
was obtained through gray scale treatment and adaptive threshold segmentation from HSV and PIV data, by which
the coordinate of each particle was marked. On the basis, the movement trajectory of each particle during
combustion was pursued by the most recent guidelines algorithm of cancroids matching. Through the method
proposed in this study, the velocity variation of each particle was obtained, the approximate distribution of particle
quantity at each zone was visualized and the mathematical model of pyrotechnic particle velocity flow field was
established.
No-reference quality assessment based on visual perception
Show abstract
The visual quality assessment of images/videos is an ongoing hot research topic, which has become
more and more important for numerous image and video processing applications with the rapid development of digital
imaging and communication technologies. The goal of image quality assessment (IQA) algorithms is to automatically
assess the quality of images/videos in agreement with human quality judgments. Up to now, two kinds of models have
been used for IQA, namely full-reference (FR) and no-reference (NR) models. For FR models, IQA algorithms interpret
image quality as fidelity or similarity with a perfect image in some perceptual space. However, the reference image is not
available in many practical applications, and a NR IQA approach is desired. Considering natural vision as optimized by
the millions of years of evolutionary pressure, many methods attempt to achieve consistency in quality prediction by
modeling salient physiological and psychological features of the human visual system (HVS). To reach this goal,
researchers try to simulate HVS with image sparsity coding and supervised machine learning, which are two main
features of HVS. A typical HVS captures the scenes by sparsity coding, and uses experienced knowledge to apperceive
objects. In this paper, we propose a novel IQA approach based on visual perception. Firstly, a standard model of HVS is
studied and analyzed, and the sparse representation of image is accomplished with the model; and then, the mapping
correlation between sparse codes and subjective quality scores is trained with the regression technique of least squaresupport
vector machine (LS-SVM), which gains the regressor that can predict the image quality; the visual metric of
image is predicted with the trained regressor at last. We validate the performance of proposed approach on Laboratory
for Image and Video Engineering (LIVE) database, the specific contents of the type of distortions present in the database
are: 227 images of JPEG2000, 233 images of JPEG, 174 images of White Noise, 174 images of Gaussian Blur, 174
images of Fast Fading. The database includes subjective differential mean opinion score (DMOS) for each image. The
experimental results show that the proposed approach not only can assess many kinds of distorted images quality, but
also exhibits a superior accuracy and monotonicity.
A position and attitude vision measurement system for wind tunnel slender model
Show abstract
A position and attitude vision measurement system for drop test slender model in wind tunnel is designed and developed.
The system used two high speed cameras, one is put to the side of the model and another is put to the position where the
camera can look up the model. Simple symbols are set on the model. The main idea of the system is based on image
matching technique between the 3D-digital model projection image and the image captured by the camera. At first, we
evaluate the pitch angles, the roll angles and the position of the centroid of a model through recognizing symbols in the
images captured by the side camera. And then, based on the evaluated attitude info, giving a series of yaw angles, a series
of projection images of the 3D-digital model are obtained. Finally, these projection images are matched with the image
which captured by the looking up camera, and the best match’s projection images corresponds to the yaw angle is the
very yaw angle of the model. Simulation experiments are conducted and the results show that the maximal error of
attitude measurement is less than 0.05°, which can meet the demand of test in wind tunnel.
An optimized fast image resizing method based on content-aware
Show abstract
In traditional image resizing theory based on interpolation, the prominent object may cause distortion, and the image
resizing method based on content-aware has become a research focus in image processing because the prominent content
and structural features of images are considered in this method. In this paper, we present an optimized fast image
resizing method based on content-aware. Firstly, an appropriate energy function model is constructed on the basis of
image meshes, and multiple energy constraint templates are established. In addition, this paper deducts the image
saliency constraints, and then the problem of image resizing is used to reformulate a kind of convex quadratic program
task. Secondly, a method based on neural network is presented in solving the problem of convex quadratic program. The
corresponding neural network model is constructed; moreover, some sufficient conditions of the neural network stability
are given. Compared with the traditional numerical algorithm such as iterative method, the neural network method is
essentially parallel and distributed, which can expedite the calculation speed. Finally, the effects of image resizing by the
proposed method and traditional image resizing method based on interpolation are compared by adopting MATLAB
software. Experiment results show that this method has a higher performance of identifying the prominent object, and the
prominent features can be preserved effectively after the image is resized. It also has the advantages of high portability
and good real-time performance with low visual distortion.
IR and visual image registration based on mutual information and PSO-Powell algorithm
Show abstract
Infrared and visual image registration has a wide application in the fields of remote sensing and military. Mutual
information (MI) has proved effective and successful in infrared and visual image registration process. To find the most
appropriate registration parameters, optimal algorithms, such as Particle Swarm Optimization (PSO) algorithm or Powell
search method, are often used. The PSO algorithm has strong global search ability and search speed is fast at the beginning,
while the weakness is low search performance in late search stage. In image registration process, it often takes a lot of time to do useless search and solution’s precision is low. Powell search method has strong local search ability. However, the search performance and time is more sensitive to initial values. In image registration, it is often obstructed by local
maximum and gets wrong results. In this paper, a novel hybrid algorithm, which combined PSO algorithm and Powell search method, is proposed. It combines both advantages that avoiding obstruction caused by local maximum and having higher precision. Firstly, using PSO algorithm gets a registration parameter which is close to global minimum. Based on the result in last stage, the Powell search method is used to find more precision registration parameter. The experimental result shows that the algorithm can effectively correct the scale, rotation and translation additional optimal algorithm. It can be a
good solution to register infrared difference of two images and has a greater performance on time and precision than
traditional and visible images.
A hyperspectral anomaly detection algorithm based on orthogonal subspace projection
Show abstract
The orithogonal subspace projection (OSP) method needs all the endmember spectral information of observation area
which is usually unavailable in actual situation. In order to extend the application of OSP method, this paper proposes an
algorithm without any priori information based on OSP. Firstly, the background endmember spectral matrix is obtained by
using unsupervised method. Then, the OSP projection operator is calculated with the background endmember matrix.
Finally, the detection operator is constructed by using the projection operator, which is used to detect the hyperspectral
imagery pixel by pixel. In order to increase the detection rate, local processing is proposed for anomaly detection with no
prior knowledge. The algorithm is tested with AVIRIS hyperspectral data, and binary image of targets and ROC curves are
given in the paper. Experimental results show that the proposed anomaly detection method based on OSP is more effective
than the classic RX detection algorithm under the case of insufficient prior knowledge, and the detection rate is
significantly increased by using the local processing.
Spatial-temporal filtering method based on kernel density estimation in suppressing background clutter
Show abstract
A temporal-spatial filtering algorithm based on kernel density estimation structure is presented for background suppression
in this paper. The algorithm can be divided into spatial filtering and temporal filtering. Smoothing process is applied to the
background of an infrared image sequence by using the kernel density estimation algorithm in spatial filtering. The
probability density of the image gray values after spatial filtering is calculated with the kernel density estimation algorithm
in temporal filtering. The background residual and blind pixels are picked out based on their gray values, and are further
filtered. The algorithm is validated with a real infrared image sequence. The image sequence is processed by using Fuller
kernel filter, Uniform kernel filter and high-pass filter. Quantitatively analysis shows that the temporal-spatial filtering
algorithm based on the nonparametric method is a satisfactory way to suppress background clutter in infrared images. The
SNR is significantly improved as well.
An adaptive point tracking method based on depth map for 2D-3D video conversion
Show abstract
The three-dimensional (3D) display technology has made a great progress in the last several decades, which provides
a dramatic improvement in visual experiences. The availability of 3D content is a critical factor limiting wide
applications of 3D technology. An adaptive point tracking method based on the depth map is demonstrated, which is used
to automatically generate depth maps elaborately. Point tracking method used in the previous investigation is template
matching and it can’t track points precisely. An adaptive point tracking method with adaptive window and weights based
on the discontinuous edge information and texture complexity of the depth map is used. In the experiment, a method to
automatically generate the depth maps using trace points between adjacent images is realized. Theoretical analysis and
experimental results show that the presented method can track feature points precisely, and the depth maps of non-key
images are perfectly generated.
The optimization of improved mean shift object tracking in embedded multicore DSP parallel system
Show abstract
This paper proposes a more robust and efficient Mean Shift object tracking algorithm which is optimized for
embedded multicore DSP Parallel system. Firstly, the RGB image is transformed into HSV image which is robust in
many aspects such as lighting changes. Then, the color histogram model is used in the back projection process to
generate the color probability distribution. Secondly, the size and position of search window are initialized in the first
frame, and Mean Shift algorithm calculates the center position of the target and adjusts the search window automatically
both in size and location, according to the result of the previous frame. Finally, since the multicore DSP system is
commonly adopted in the embedded application such as seeker and an optical scout system, we implement the proposed
algorithm in the TI multicore DSP system to meet the need of large amount computation. For multicore parallel
computing, the explicit IPC based multicore framework is designed which outperforms OpenMP standard. Moreover,
the parallelisms of 8 functional units and cross path data fetch capability of C66 core are utilized to accelerate the
computation of iteration in Mean Shift algorithm. The experimental results show that the algorithm has good
performance in complex scenes such as deformation, scale change and occlusion, simultaneously the proposed
optimization method can significantly reduce the computation time.
Degradation and restoration of high resolution TDICCD imagery due to satellite vibrations
Show abstract
A new method is proposed to solve the problem of image restoration of high resolution TDICCD camera due to satellite
vibrations, which considers image blur and irregular sampling geometric quality degradation simultaneously. Firstly, the
image quality degradation process is analyzed according to imaging characteristics of TDICCD camera, owing to image
motions during TDICCD integration time induced by satellite vibrations. In addition, the vibration simulation model is
established, and the irregular sampling degradation process of geometric quality is mathematically modeled using
bicubic Hermite interpolation. Subsequently, a full image degradation model is developed combined with blurred and
noisy degradation process. On this basis, a new method of image restoration is presented, which can implement not only
deblurring but also irregular to regular sampling. Finally, the method is verified using real remote sensing images, and
compared with the recent restoration methods. Experimental results indicate that the proposed method performs well,
and the Structural Similarity between the restored and ideal images are greater than 0.9 in the case of seriously blurred,
irregularly sampled and noisy images. The proposed method can be applied to restore high resolution on-orbit satellite
images effectively.
Identification method of satellite local components based on combined feature metrics
Show abstract
In order to meet the requirements of identification of satellite local targets, a new method based on combined feature
metrics is proposed. Firstly, the geometric features of satellite local targets including body, solar panel and antenna are
analyzed respectively, and then the cluster of each component are constructed based on the combined feature metrics of
mathematical morphology. Then the corresponding fractal clustering criterions are given. A cluster model is established,
which determines the component classification according to weighted combination of the fractal geometric features. On
this basis, the identified targets in the satellite image can be recognized by computing the matching probabilities between
the identified targets and the clustered ones, which are weighted combinations of the component fractal feature metrics
defined in the model. Moreover, the weights are iteratively selected through particle swarm optimization to promote
recognition accuracy. Finally, the performance of the identification algorithm is analyzed and verified. Experimental
results indicate that the algorithm is able to identify the satellite body, solar panel and antenna accurately with
identification probability up to 95%, and has high computing efficiency. The proposed method can be applied to identify
on-orbit satellite local targets and possesses potential application prospects on spatial target detection and identification.
Multiple dim targets detection in infrared image sequences
Show abstract
Strong noises interference is a difficult technical problem for signals detection. Multiple targets detection with strong
noises makes the problem more complicated. Aiming at the difficulty of multiple uniform rectilinear motion targets
detection in infrared (IR) image sequences with strong noises, this paper presents a multiple dim targets detection
algorithm which improves signal-to-noise ratio (SNR). Firstly, we establish a velocity space and stack image sequences
along different velocity vectors. Secondly, mean filtering in time-domain is applied to stacked images. Thirdly,
quasi-target points in mean filtering images are selected by constant false-alarm ratio (CFAR) judging. Finally,
coordinate vectors and velocity vectors of quasi-target points are mapped to location space and velocity space,
respectively. As a result, local peaks from the two spaces will confirm target points; meanwhile, velocity vectors of
targets can also be acquired. In addition, effect of velocity steps on SNR improvement is analyzed, which can guide the
selection of steps and reduce computational burden. Both moving dim targets simulation experiment and real-world dim
targets detection experiment have proved that this algorithm can effectively detect multiple dim targets under strong
noise background.
A novel iris patterns matching algorithm of weighted polar frequency correlation
Weijie Zhao,
Linhua Jiang
Show abstract
Iris recognition is recognized as one of the most accurate techniques for biometric authentication. In this paper,
we present a novel correlation method - Weighted Polar Frequency Correlation(WPFC) - to match and evaluate
two iris images, actually it can also be used for evaluating the similarity of any two images. The WPFC
method is a novel matching and evaluating method for iris image matching, which is complete different from
the conventional methods. For instance, the classical John Daugman’s method of iris recognition uses 2D Gabor
wavelets to extract features of iris image into a compact bit stream, and then matching two bit streams with
hamming distance. Our new method is based on the correlation in the polar coordinate system in frequency
domain with regulated weights. The new method is motivated by the observation that the pattern of iris that
contains far more information for recognition is fine structure at high frequency other than the gross shapes of
iris images. Therefore, we transform iris images into frequency domain and set different weights to frequencies.
Then calculate the correlation of two iris images in frequency domain. We evaluate the iris images by summing
the discrete correlation values with regulated weights, comparing the value with preset threshold to tell whether
these two iris images are captured from the same person or not. Experiments are carried out on both CASIA
database and self-obtained images. The results show that our method is functional and reliable. Our method
provides a new prospect for iris recognition system.
Projector calibration method based on optical coaxial camera
Show abstract
This paper presents a novel method to accurately calibrate a DLP projector by using an optical coaxial camera to capture
the needed images. A plate beam splitter is used to make imaging axis of the CCD camera and projecting axis of the DLP
projector coaxial, so the DLP projector can be treated as a true inverse camera. A plate having discrete markers on the
surface will be designed and manufactured to calibrate the DLP projector. By projecting vertical and horizontal
sinusoidal fringe patterns on the plate surface from the projector, the absolute phase of each marker’s center can be
obtained. The corresponding projector pixel coordinate of each marker is determined from the obtained absolute phase.
The internal and external parameters of the DLP projector are calibrated by the corresponding point pair between the
projector coordinate and the world coordinate of discrete markers. Experimental results show that the proposed method
accurately obtains the parameters of the DLP projector. One advantage of the method is the calibrated internal and
external parameters have high accuracy because of uncalibrating the camera. The other is the optical coaxes geometry
gives a true inverse camera, so the calibrated parameters are more accurate than that of crossed-optical-axes, especially
the principal points and the radial distortion coefficients of the projector lens.
The electromagnetic simulation of radar imaging of complicated satellites
Wenjing Chen,
Chunzhu Dong,
Chaoying Huo,
et al.
Show abstract
Research on electromagnetic scattering from electrical large space target, random rough surface, and the composite model of space target and rough surface, has been more and more important in recent years. This paper presents studies of geometrical modeling, simulation of rough surface of satellites and analyzing of radar satellite image from scattering phenomenology. The Gaussian random fluctuation is adopted in the electromagnetic compute to simulate diffuse reflectance caused by rough surface of satellite. The efficient and accurate simulation of complicated satellites is realizable. Wide-band electromagnetic scattering characteristics which are obtained by this method could be used to analyze the information of structure and shape of satellites more accurately. It is important for imagery interpretation of space targets.
Grassmann manifold based shape matching and retrieval under partial occlusions
Show abstract
Shape matching and recognition is a challenging task due to geometric distortions and occlusions. A novel shape
matching approach based on Grassmann manifold is proposed that affine transformations and partial occlusions are both
considered. An affine invariant Grassmann shape descriptor is employed which projects one shape contour to a point on
Grassmann manifold and gives the similarity measurement between two contours based on the geodesic distance on the
manifold. At first, shape contours are parameterized by affine length and accordingly divided into local affine-invariant
shape segments, which are represented by the Grassmann shape descriptor, according to their curvature scale space
images. Then the Smith-Waterman algorithm is employed to find the common parts of two shapes’ segment sequences,
and get the local similarity of shapes. The global similarity is given by the found common parts, and finally the shape
recognition accomplished by the weighted sum of local similarity and global similarity. The robustness of the Grassmann
shape descriptor is analyzed through subspace perturbation analysis theory. Retrieval experiments show that our
approach is effective and robust under affine transformations and partial occlusions.
Visual fatigue modeling for stereoscopic video shot based on camera motion
Show abstract
As three–dimensional television (3-DTV) and 3-D movie become popular, the discomfort of visual feeling limits
further applications of 3D display technology. The cause of visual discomfort from stereoscopic video conflicts between
accommodation and convergence, excessive binocular parallax, fast motion of objects and so on. Here, a novel method
for evaluating visual fatigue is demonstrated. Influence factors including spatial structure, motion scale and comfortable
zone are analyzed. According to the human visual system (HVS), people only need to converge their eyes to the specific
objects for static cameras and background. Relative motion should be considered for different camera conditions
determining different factor coefficients and weights. Compared with the traditional visual fatigue prediction model, a
novel visual fatigue predicting model is presented. Visual fatigue degree is predicted using multiple linear regression
method combining with the subjective evaluation. Consequently, each factor can reflect the characteristics of the scene,
and the total visual fatigue score can be indicated according to the proposed algorithm. Compared with conventional
algorithms which ignored the status of the camera, our approach exhibits reliable performance in terms of correlation
with subjective test results.
A detection method of moving object based on hybrid difference
Show abstract
The detection method is based on background subtraction and inter-frame difference. To use statistical model of RGB color histograms to extracting background. In this way, the initial background image could be extracted without noise effect to a great extent. To get difference image of moving object according to the results of background subtraction and three frames difference. To get binary Image A which difference from Frame k-1 and Frame k, to get Image B which difference from Frame k and Frame k+1. Let Image A and Image B do LOR operation to get Image C for obtaining more information of the moving object. Finally, let binary image of background subtraction and Image C do LAND operation to get outline of moving object. To use self-adaption method updates background image to promise the instantaneity. If a pixel of the current frame is estimated as moving target, we set the corresponding pixel of current background image to instead of the pixel in background image, else set the corresponding pixel of current frame to update the corresponding pixel of background. To use background updating factor α to control update rate. Moving object can be detected more accurately by mathematical morphology. This method can improve the shortcomings of background subtraction and inter-frame difference.
An automatic registration algorithm of infrared and visible images based on hybrid image features
Show abstract
In this paper, a new method is presented to match a pair of visible and infrared images of same scene based on hybrid visual features including line segments and interest points. First, improved Harris corner extraction method and line segment detector method is used to extract feature points and segments. Then, a novel descriptor integrating the information of line segments and interest points is proposed. Finally, the nearest neighbor algorithm is utilized to match the descriptors, and the RANSAC(Random Sample Consensus) algorithm is employed to rule out the wrong match pairs. The performances are evaluated by extensive experiments on real images. The results show that the proposed algorithm can take advantage of similar structures between the multimodal images to realize automatic matching efficiently.
Compression imaging based on Fourier transform optical system
Show abstract
Shannon / Nyquist sampling theorem indicates that during the sampling process the minimum sample rate must be more than the double of the band of the signal so that we can achieve images without distortion. High-frequency sampling leads to mass data and results in high cost of storage and transmission procedure. Compressed sensing indicates that we can sample data at far below the Nyquist frequency when the signals are sparse or can be represented as sparse on some orthogonal basis, and the signals can be recovered without distortion after some certain recovery algorithms. By this means the cost of storage and transmission can be reduced significantly. Unlike conventional optical imaging process, this paper presents a new imaging method using a Fourier transform lens system, which enables single-exposure and single-aperture compressed imaging. First, the Fourier transformation of image signals is accomplished after they pass through a Fourier transform optical system. Second, sparse sample data can be obtained after the spectrum signals pass the sensor array. The process mentioned above can be interpreted as that using a Fourier matrix and a sparse matrix to complete the measurement of the image signals. Third, we make use of fast iterative threshold recovery algorithm to compute the sampling values and obtain the target image signals. Compared with the conventional imaging methods, in the case of ensuring the image quality, our method can significantly reduce the number of samples, thus greatly reduce the data redundancy. Simulation results indicate that the imaging method proposed can be prospective.
Robust visual tracking of infrared object via sparse representation model
Show abstract
In this paper, we propose a robust tracking method for infrared object. We introduce the appearance model and the sparse representation in the framework of particle filter to achieve this goal. Representing every candidate image patch as a linear combination of bases in the subspace which is spanned by the target templates is the mechanism behind this method. The natural property, that if the candidate image patch is the target so the coefficient vector must be sparse, can ensure our algorithm successfully. Firstly, the target must be indicated manually in the first frame of the video, then construct the dictionary using the appearance model of the target templates. Secondly, the candidate image patches are selected in following frames and the sparse coefficient vectors of them are calculated via ℓ1-norm minimization algorithm. According to the sparse coefficient vectors the right candidates is determined as the target. Finally, the target templates update dynamically to cope with appearance change in the tracking process. This paper also addresses the problem of scale changing and the rotation of the target occurring in tracking. Theoretic analysis and experimental results show that the proposed algorithm is effective and robust.
Non-uniformity correction for sequence-image based on local histogram specification
Show abstract
Non-uniformity noise is one of the critical problems that need to be solved in imaging system. Due to the non-uniform of each detector, there are regular strip noises in the image, which has serious effect on the image quality and thus on image applications. In target recognition system, there is also intensity difference between the adjacent images in the sequencial images other than the uniformity between each image. In this paper, a new non-uniformity correction method for sequence-image based on local histogram specification is proposed. The process is aimed to suppress the non-uniformity noise and smooth the image. The two adjacent images are interlaced into one image column by column. And then, the image is processed by the local weight histogram specification method . Experimental results show that the method presented in this paper can effectively suppress the strip noise in each image and thus reduce the gray level difference among the image sequences.
Fast small target tracking in IR imagery based on improved similarity measure
Show abstract
In order to enhance the robustness of IR fast small target tracking, a novel mean shift tracking algorithm using improved similarity measure of is proposed. Firstly, problems of local background interfering in original mean shift algorithm for tracking fast motion small target is analyzed, and the reasons is located in the Bhattacharyya coefficient similarity measure expression for all gray weights of components are same, which cannot reflect the advantage contribution of the small target’s gray component in the process of measuring similarity, causing serious interference of the background in the tracking process, leaving the algorithm converging easily. Therefore, to solve this problem, the improvements Bhattacharyya coefficient similarity measure with the local background information fused is proposed. Then, shift vector is deduced in the framework of mean shift by regarding Bhattacharyya coefficients as the similarity measure.In shifting process, the robustness of the small target tracking is improved effectively according to target gray level of large membership degree with high shift weight, and vice versa with low shift weight, which the background interference is suppressed to some extent. In sake of verifying the performance of the proposed algorithm, the classical mean shift algorithm and the algorithm of this paper is used in the target tracking simulation experiment, as well as the infrared image sequences containing the small fast targets of uncooled infrared camera is used. Finally the experimental result indicates that the performance of tracking the small fast target in IR images is robust.
Rectangle object segmentation based on shape preserving and CV variational level set
Show abstract
In order to solve the problems that CV model can’t segment object which is partially occluded or has similar gray value with background or has obvious textures, we add shape restraint equations of prior shape to level set function, which keeps the curve to be a specific class shape in the whole evolvement, thus we realize shape preserving in object segmentation. In addition, we build an energy function for rectangle object using our proposed model, deduce a group of corresponding Euler-Lagrange ordinary differential functions and evolve the level set function. By evolution, rectangle object can be segmented, and the final level set function is just the quantitative description of the rectangle object. At last, we validate with three groups of experiments that our model can not only segment the rectangle object from complex backgrounds, but also has lessened calculation and strong robustness.
The speckle image reconstruction of the solar small scale features
Show abstract
The resolution of the astronomical object observed by the earth-based telescope is limited due to the atmospheric turbulence. Speckle image reconstruction method provides access to detect small-scale solar features near the diffraction limit of the telescope. This paper describes the implementation of the reconstruction of images obtained by the 1-m new vacuum solar telescope at Full-Shine solar observatory. Speckle masking method is used to reconstruct the Fourier phases for its better dynamic range and resolution capabilities. Except of the phase reconstruction process, several problems encounter in the solar image reconstruction are discussed. The details of the implement including the flat-field, image segmentation, Fried parameter estimation and noise filter estimating are described particularly. It is demonstrated that the speckle image reconstruction is effective to restore the wide field of view images. The qualities of the restorations are evaluated by the contrast ratio. When the Fried parameter is 10cm, the contrast ratio of the sunspot and granulation can be improved from 0.3916 to 0.6845 and from 0.0248 to 0.0756 respectively.
A multi-sensor image fusion algorithm based on region-based selection and IHS transform
Show abstract
The multi-sensor image fusion technology can obtain a more comprehensive and more accurate and reliable image, in
order to understand the scene or recognize the target more easily. However, most existing algorithms are mainly based on
optical remote sensing images, which is highly susceptible by media interference, supplemented by SAR images. The
image fusion between SAR images and PAN images also cannot save the textural feature and the color information
effectively at the same time. In view of these problems, this paper presents a multi-sensor image fusion algorithm based
on region-based selection and IHS transform. The SAR image and PAN image are firstly IHS transformed to achieve the
intensity (I), hue (H) and saturation (S) weights. The I weights of SAR image and PAN image are separately decomposed
using SIDWT algorithm to extract wavelet coefficients. Then, the I weight of SAR image is divided into regular area and
irregular area based on a new adaptive segmentation method. A new fusion rules is presented according to local feature,
and then used to fuse corresponding wavelet coefficients of the I weight of SAR image and PAN image. Inverse SIDWT
is carried out on the fused wavelet coefficients to get the I weight (I’) of fused image. Finally, the fused image is obtained
by inverse IHS transform of I’ weight with the H, S weight of PAN image. Experimental results of real images validated
the effectiveness of the proposed algorithm by objective evaluation such as standard deviation, entropy, average gradient,
etc.
Robust and fast license plate detection based on the fusion of color and edge feature
Show abstract
Extracting a license plate is an important stage in automatic vehicle identification. The degradation of images and the computation intense make this task difficult. In this paper, a robust and fast license plate detection based on the fusion of color and edge feature is proposed. Based on the dichromatic reflection model, two new color ratios computed from the RGB color model are introduced and proved to be two color invariants. The global color feature extracted by the new color invariants improves the method’s robustness. The local Sobel edge feature guarantees the method’s accuracy. In the experiment, the detection performance is good. The detection results show that this paper’s method is robust to the illumination, object geometry and the disturbance around the license plates. The method can also detect license plates when the color of the car body is the same as the color of the plates. The processing time for image size of 1000x1000 by pixels is nearly 0.2s. Based on the comparison, the performance of the new ratios is comparable to the common used HSI color model.
A simple and efficient object detection method based on saliency measure for infrared radiation image
Show abstract
Detection of visually salient objects plays an important role in applications such as object segmentation, adaptive compression, object recognition, etc. A simple and computationally efficient method is presented in this paper for detecting visually salient objects in Infrared Radiation images. The proposed method can be divided into three steps. Firstly, the infrared image is pre-processed to increase the contrast between objects and background. Secondly, the spectral residual of the pre-processed image is extracted in the log spectrum, then via corresponding inverse transform and threshold segmentation we can get the rough regions of the salient objects. Finally, we apply a sliding window to acquire the explicit position of the salient objects using the probabilistic interpretation of the semi-local feature contrast which is estimated by comparing the gray level distribution of the object and the surrounding area in the original image. And as we change the size of the sliding window, different size of objects can be found out. In our proposed method, the first two steps combined together to play a role in narrowing the searching region and thus accelerating computation. The third procedure is applied to extract the salient objects. We test our method on abundant amount of Infrared Radiation images, and the results show that our saliency detection based object detection method is effective and robust.
A method of plane geometry primitive presentation
Show abstract
Point feature and line feature are basic elements in object feature sets, and they play an important role in object matching and recognition. On one hand, point feature is sensitive to noise; on the other hand, there are usually a huge number of point features in an image, which makes it complex for matching. Line feature includes straight line segment and curve. One difficulty in straight line segment matching is the uncertainty of endpoint location, the other is straight line segment fracture problem or short straight line segments joined to form long straight line segment. While for the curve, in addition to the above problems, there is another difficulty in how to quantitatively describe the shape difference between curves. Due to the problems of point feature and line feature, the robustness and accuracy of target description will be affected; in this case, a method of plane geometry primitive presentation is proposed to describe the significant structure of an object. Firstly, two types of primitives are constructed, they are intersecting line primitive and blob primitive. Secondly, a line segment detector (LSD) is applied to detect line segment, and then intersecting line primitive is extracted. Finally, robustness and accuracy of the plane geometry primitive presentation method is studied. This method has a good ability to obtain structural information of the object, even if there is rotation or scale change of the object in the image. Experimental results verify the robustness and accuracy of this method.
Human action recognition based on spatial-temporal descriptors using key poses
Show abstract
Human action recognition is an important area of pattern recognition today due to its direct application and need in various occasions like surveillance and virtual reality. In this paper, a simple and effective human action recognition method is presented based on the key poses of human silhouette and the spatio-temporal feature. Firstly, the contour points of human silhouette have been gotten, and the key poses are learned by means of K-means clustering based on the Euclidean distance between each contour point and the centre point of the human silhouette, and then the type of each action is labeled for further match. Secondly, we obtain the trajectories of centre point of each frame, and create a spatio-temporal feature value represented by W to describe the motion direction and speed of each action. The value W contains the information of location and temporal order of each point on the trajectories. Finally, the matching stage is performed by comparing the key poses and W between training sequences and test sequences, the nearest neighbor sequences is found and its label supplied the final result. Experiments on the public available Weizmann datasets show the proposed method can improve accuracy by distinguishing amphibious poses and increase suitability for real-time applications by reducing the computational cost.
Traffic flow visualization based on line integral convolution
Show abstract
Traffic flow visualization is an important tack in traffic management and computer vision field. Traditional methods use the velocities of particles of the moving vehicles such as optical flow to visualize the traffic flow. However, using optical flow can only gain a coarse description of traffic flow. Many details in the flow field are missed. Texture synthesizing technology is a suitable tool for flow field visualization, which can represent the flow field as a texture image. This paper proposed a visualization method to represent traffic flow as a texture image. Firstly, Horn-Schunck optical flow is calculated between two consecutive frames. In order to reveal more details of a traffic flow field, Line Integral Convolution (LIC) is used by convolute noise texture along the streamline of the optical flow field. Therefore, the moving vehicles can be represented as a texture images. On the contrary, the background regions are mapped as noise. Experimental results show the proposed method can show the traffic flow clearer than optical flow.
Long-term object tracking combined offline with online learning
Show abstract
In this paper we propose a simply yet effective and efficient method for long-term object tracking. Different from traditional visual tracking method which mainly depends on frame-to-frame correspondence, we combine high-level semantic information with low-level correspondences. Our framework is formulated in a confidence selection framework, which allows our system to recover from drift and partly deal with occlusion problem. To summarize, our algorithm can be roughly decomposed in a initialization stage and a tracking stage. In the initialization stage, an offline classifier is trained to get the object appearance information in category level. When the video stream is coming, the pre-trained offline classifier is used for detecting the potential target and initializing the tracking stage. In the tracking stage, it consists of three parts which are online tracking part, offline tracking part and confidence judgment part. Online tracking part captures the specific target appearance information while detection part localizes the object based on the pre-trained offline classifier. Since there is no data dependence between online tracking and offline detection, these two parts are running in parallel to significantly improve the processing speed. A confidence selection mechanism is proposed to optimize the object location. Besides, we also propose a simple mechanism to judge the absence of the object. If the target is lost, the pre-trained offline classifier is utilized to re-initialize the whole algorithm as long as the target is re-located. During experiment, we evaluate our method on several challenging video sequences and demonstrate competitive results.
Boat detection using vector accumulation of particle motion
Show abstract
Recently, target detection in sea environment such as boat detection has become a popular research topic which is significant for marine vessels monitoring system. Many target detection methods have been widely applied to practical applications such as frame difference, traditional optical flow and background subtraction method. However, the existing target detection methods are not suitable to deal with the complex conditions of sea surface, such as irregular movement of the waves and illumination changes. In this paper, we developed an approach based on vector accumulation of particle motion mainly aiming at eliminating the effects of irregular movement of waves. Our proposed method applies vector accumulation of particle motion to optical flow field to obtain more accurate detection results under complex conditions. Firstly, the traditional optical flow method is used to acquire motion vector of every particle. Furthermore, the vectors of each flow point are abstracted to represent the recording of a fluid element in the flow over a certain period, succeeding is the accumulation of particle vectors. Finally, we calculate the mean of the vector accumulation to eliminate the effects of irregular movement of waves based on the video. Experimental results show the proposed method can gain better performance than traditional optical flow method.
Fast object reconstruction in block-based compressive low-light-level imaging
Show abstract
In this paper we propose a simply yet effective and efficient method for long-term object tracking. Different from traditional visual tracking method which mainly depends on frame-to-frame correspondence, we combine high-level semantic information with low-level correspondences. Our framework is formulated in a confidence selection framework, which allows our system to recover from drift and partly deal with occlusion problem. To summarize, our algorithm can be roughly decomposed in a initialization stage and a tracking stage. In the initialization stage, an offline classifier is trained to get the object appearance information in category level. When the video stream is coming, the pre-trained offline classifier is used for detecting the potential target and initializing the tracking stage. In the tracking stage, it consists of three parts which are online tracking part, offline tracking part and confidence judgment part. Online tracking part captures the specific target appearance information while detection part localizes the object based on the pre-trained offline classifier. Since there is no data dependence between online tracking and offline detection, these two parts are running in parallel to significantly improve the processing speed. A confidence selection mechanism is proposed to optimize the object location. Besides, we also propose a simple mechanism to judge the absence of the object. If the target is lost, the pre-trained offline classifier is utilized to re-initialize the whole algorithm as long as the target is re-located. During experiment, we evaluate our method on several challenging video sequences and demonstrate competitive results.
GPU-based parallel optimization implement of phase diversity
Show abstract
Phase diversity (PD) can not only be used as wavefront sensor but also as image post processing technique. However, its computations have been perceived as being too burdensome and it is difficult to achieve its real time application on a PC platform. In this paper, we carried out parallel analysis on the algorithm and task assignments on the heterogeneous platform of CPU-GPU, and then implement parallel programing optimization on GPUs. The optimization strategies of the cost function on GPU are introduced. The process of OTF is improved to make the amount of calcuation reduced by 11% compared to the original method. In order to demonstrate the speedup of PD, two images, 128x128 pixels and 256x256 pixels in dimension, are tested on CPU platform and CPU/GPU heterogeneous platform respectively. The results show the time costs have the improvenments of 13x and 28x for the implementation of PD based on GPU in contrast with that based on CPU.
Point target enhancement based on temporal-spatial over-sampling and image filtering
Show abstract
A method for point target enhancement based on temporal-spatial over-sampling and adaptive filtering is proposed in this paper. First of all, an over-sampling scanning imaging system is designed for target imaging enhancement. Two separate detector arrays are offset to each other by a half detector in the cross-scan direction and the sampling frequency in the in-scan direction is increased. A sub-pixel image of point target is obtained by interlace-combined two frame image from the two detector arrays. Secondly, image filtering is used to enhance the target embedded in clutter background. Clutter background suppression, local enhancement and contrast extension are performed. Experimental results show that the method presented in this paper can enhance the point target effectively in scanning detection system.
Design of automatic image measuring system based on RTX simulation system
Show abstract
This paper focuses on the embedded laser target simulation system based on RTX(Real Time Executive).This system completes the seeker test mission mainly by means of simulating the process of the missile from launch to hit the target through controlling the Laser spot. The system is consisted of upper computer, spot energy adjustment and spot size adjustment. The whole system realizes continuous and rapid adjustment to the energy and size of the spot via the RTX embedded computer system sends control commands to each controller of two units simultaneously. The unit of spot size adjustment: the controller controls the movement of the leading screw and then the leading screw controls the shift of the lens. The motor which controls the movement of the leading screw is direct current torque motor. The unit of spot energy adjustment: According to the Lambert law, the structure uses two crystals that the wedge angles are exactly the same. The system controls the motor to change the relative position of these crystals and then changes the thickness of the optical path crystal to adjust the laser spot energy. This system utilizes RTX embedded system to send control commends s in real time, a share memory area is opened for exchanging the simultaneous data between Windows and RTX. The real-time performance of the whole system has a significantly improve by using RTX embedded system, the system response time reduces to 1ms。Thus the laser target achieves rapid changing which seeker test requirement.
Adaptive infrared-image details enhancement technology
Show abstract
In order to surmount the infrared-image object differentiation difficulty caused by the blurred image edge, a kind of adaptive filter based infrared-image nonlinear edge enhancement technology was proposed in this paper. This technology integrates image nonlinear edge-sharpening and Multi-scale analyze method. The approach of Gauss pyramid structure can enhance detail information by using non-linear algorithms in different scales. The enhanced detail information is then added back to the original image iteratively. While saving the image edge information it can filter image noise and edge distortion caused by edge-sharpening and improve image’s clarity and SNR obviously. Gray scale grads was defined based on gray linear increment, image edge enhancement arithmetic can be real time realized, and has been applied in high performance thermal imager. As it is shown in experiments, this algorithm has practicality and potential application value in the field of infrared images contrast enhancement
Robust detection of key point of a building for infrared imaging sequence
Show abstract
In this paper, we describe a fully automatic approach for detecting and matching geometrical corner feature correspondences between aerial images with larger scale and view variations. The main assumption of the approach is the fact that many man-made environments contain a large number of parallel linear features. We exploit this observation towards efficient detection and estimation of vanishing points. Given the vanishing points within an image, building geometrical corner features are obtained by the intersections of pairs of building outlines corresponding to different vanishing points. The experiments performed on the infrared aerial image sequences evaluate the stability and distinctiveness of the proposed features which are undergone appearance changes due to projective deformation.
The research of infrared image enhancement algorithm based on human vision
Show abstract
Infrared images have their own characteristics: low contrast, great noise, large dynamic range and poor visual effect. The traditional image enhancement algorithms have certain limitations and can't achieve a good visual effect. In order to obtain a good visual effect and improve the target detection and recognition capabilities, the paper studied various enhancement methods. After analyzing the retinex theory, we choose the image enhancement method based on human visual system called retinex to process infrared images. Retinex has been used to enhance the visible light image. To do experiment on infrared image enhancement, multi-scale retinex method gets ideal visual effect. On this basis, we propose an improved multi-scale Retinex (AMSR) method based on adaptive adjustment. This method can adaptively adjust the gray level and contrast of the image, enhance the details, make the weak small targets more conducive to the human eye observation. While, it is impossible to find a method suited for all infrared images with different characteristics. So, we use several traditional image enhancement algorithms to compare with the retinex algorithms. And calculate the objective evaluation factors, including average, standard deviation, entropy and so on. After observation the processing results and analyzing these evaluation factors, the AMSR algorithm is proved having its applicability and superiority. In order to select a suitable infrared image enhancement algorithms, we analyze the applicability of each enhancement methods for infrared image has obvious characteristics, To some extent, the study is significant to the infrared target detection and recognition.
Center determination for trailed sources in astronomical observation images
Show abstract
Images with trailed sources can be obtained when observing near-Earth objects, such as small astroids, space debris, major planets and their satellites, no matter the telescopes track on sidereal speed or the speed of target. The low centering accuracy of these trailed sources is one of the most important sources of the astrometric uncertainty, but how to determine the central positions of the trailed sources accurately remains a significant challenge to image processing techniques, especially in the study of faint or fast moving objects. According to the conditions of one-meter telescope at Weihai Observatory of Shandong University, moment and point-spread-function (PSF) fitting were chosen to develop the image processing pipeline for space debris. The principles and the implementations of both two methods are introduced in this paper. And some simulated images containing trailed sources are analyzed with each technique. The results show that two methods are comparable to obtain the accurate central positions of trailed sources when the signal to noise (SNR) is high. But moment tends to fail for the objects with low SNR. Compared with moment, PSF fitting seems to be more robust and versatile. However, PSF fitting is quite time-consuming. Therefore, if there are enough bright stars in the field, or the high astronometric accuracy is not necessary, moment is competent. Otherwise, the combination of moment and PSF fitting is recommended.
A color fusion method of infrared and low-light-level images based on visual perception
Show abstract
The color fusion images can be obtained through the fusion of infrared and low-light-level images, which will contain both the information of the two. The fusion images can help observers to understand the multichannel images comprehensively. However, simple fusion may lose the target information due to inconspicuous targets in long-distance infrared and low-light-level images; and if targets extraction is adopted blindly, the perception of the scene information will be affected seriously. To solve this problem, a new fusion method based on visual perception is proposed in this paper. The extraction of the visual targets (“what” information) and parallel processing mechanism are applied in traditional color fusion methods. The infrared and low-light-level color fusion images are achieved based on efficient typical targets learning. Experimental results show the effectiveness of the proposed method. The fusion images achieved by our algorithm can not only improve the detection rate of targets, but also get rich natural information of the scenes.
An adaptive unsupervised hyperspectral classification method based on Gaussian distribution
Show abstract
In order to achieve adaptive unsupervised clustering in the high precision, a method using Gaussian distribution to fit the similarity of the inter-class and the noise distribution is proposed in this paper, and then the automatic segmentation threshold is determined by the fitting result. First, according with the similarity measure of the spectral curve, this method assumes that the target and the background both in Gaussian distribution, the distribution characteristics is obtained through fitting the similarity measure of minimum related windows and center pixels with Gaussian function, and then the adaptive threshold is achieved. Second, make use of the pixel minimum related windows to merge adjacent similar pixels into a picture-block, then the dimensionality reduction is completed and the non-supervised classification is realized. AVIRIS data and a set of hyperspectral data we caught are used to evaluate the performance of the proposed method. Experimental results show that the proposed algorithm not only realizes the adaptive but also outperforms K-MEANS and ISODATA on the classification accuracy, edge recognition and robustness.
3D surface reconstruction from range slices
Show abstract
Research on three-dimensional (3D) surface reconstruction from range slices obtained from range-gated laser imaging system is of significance. 3D surfaces reconstructed based on existing binarization method or centroid method are rough or discontinuous in some circumstances. In this paper we address these problems and develop a 3D surface reconstruction algorithm based on the idea that combining the centroid method with weighted linear interpolation and mean filter. The algorithm consists of three steps. In the first step, interesting regions are extracted from each range slice based on mean filter, and then are merged to derive a single range image. In the second step, the derived range image is denoised and smoothed based on adaptive histogram method, weighted linear interpolation and mean filter method respectively. Finally, nonzero valued pixels in the after processed range image are converted to point cloud according to the range-gated imaging parameters, and then 3D surface meshes are established from the point cloud based on the topological relationship between adjacent pixels in the range image. Experiment is conducted on range slices generated from range-gated laser imaging simulation platform, and the registration result of the reconstructed surface of our method with the original surface of the object shows that the proposed method can reconstruct object surface accurately, so it can be used for the designing of reconstruction and displaying of range-gated laser imaging system, and also can be used for 3D object recognition.
An ISAR imaging algorithm for the space satellite based on empirical mode decomposition theory
Show abstract
Currently, high resolution imaging of the space satellite is a popular topic in the field of radar technology. In contrast with regular targets, the satellite target often moves along with its trajectory and simultaneously its solar panel substrate changes the direction toward the sun to obtain energy. Aiming at the imaging problem, a signal separating and imaging approach based on the empirical mode decomposition (EMD) theory is proposed, and the approach can realize separating the signal of two parts in the satellite target, the main body and the solar panel substrate and imaging for the target. The simulation experimentation can demonstrate the validity of the proposed method.
Extraction of contour based on zero-crossing feature and contour measure
Show abstract
We present a method to extract edges using zero-crossing feature and contour measure. This method differs markedly from previous ones, since it provided a means of quantitative analysis to detect zero-crossing. There are two main steps in this method. Firstly, the edge intensity was obtained through the value of contour measure. Secondly, the actual edges are identified according to the edges intensity. A series of experiments are performed to test the algorithm proposed, which show that the edges is extracted more accurately and completely.
Simulation of the atmospheric turbulence image reconstruction based on compressed sensing
Show abstract
For long distance imaging in the atmosphere environment, the image quality is affected by the atmospheric turbulence. In this paper, the image quality is analyzed, while the atmospheric turbulence is simulated through the spectrum inversion method of the Fourier transform. The target images affected by both weak and strong atmospheric turbulence environments are reconstructed respectively, employing the compressed sensing algorithm. Results show that the compressed sensing algorithm inhibits the atmospheric turbulence effect to some extent, but not that brilliant in the strong turbulence conditions. Outstanding performance of the compressed sensing algorithm in image noise reduction is also confirmed.
Non-local means-based nonuniformity correction for infrared focal-plane array detectors
Hui Yu,
Zhi-jie Zhang,
Fu-sheng Chen,
et al.
Show abstract
The infrared imaging systems are normally based on the infrared focal-plane array (IRFPA) which can be considered as an array of independent detectors aligned at the focal plane of the imaging system. Unfortunately, every detector on the IRFPA may have a different response to the same input infrared signal which is known as the nonuniformity problem. Then we can observe the fixed pattern noise (FPN) from the resulting images. Standard nonuniformity correction (NUC) methods need to be recalibrated after a short period of time due the temporal drift of the FPN. Scene-based nonuniformity correction (NUC) techniques eliminate the need for calibration by correction coefficients based on the scene being viewed. However, in the scene-based NUC method the problem of ghosting artifacts widely seriously decreases the image quality, which can degrade the performance of many applications such as target detection and track. This paper proposed an improved scene-based method based on the retina-like neural network approach. The method incorporates the use of non-local means (NLM) method into the estimation of the gain and the offset of each detector. This method can not only estimates the accurate correction coefficient but also restrict the ghosting artifacts efficiently. The proposed method relies on the use of NLM method which is a very successful image denoising method. And then the NLM used here can preserve the image edges efficiently and obtain a reliable spatial estimation. We tested the proposed NUC method by applying it to an IR sequence of frames. The performance of the proposed method was compared the other well-established adaptive NUC techniques.
Sea-level line extraction based on piecewise line detection
Show abstract
In infrared image, sea-level line could be hard to distinguish because of noises caused by wave clutters and sunlight conditions.This paper proposed a fast sea-level line extraction method which could localize the sea-level line in complex infrared sea-sky scenes. First, a down sample operation was performed to obtain a low resolution image which could reduce the time consumption without blurring the sea-level line, and then the Canny edge detection was carried out to extract edges in the scene. Second, the intersecting edges were separated by removing the joints of edges according to a certain rule, and the bounding rectangle of every short edge was obtained which helped to select straight lines, and then a long edge segmentation operation was used to count in possible sea-level line. Third, a line concatenation method was performed by their slopes and intercepts comparison. Finally, for sea-level line verification, the second-order vertical grads are calculated in the two sides of possible sea-level line. Experiments show that the proposed method is fast and effective for various kinds of infrared sea-sky scenes, and it is feasible even for the scenes where the sea-level line is blurring and hard to distinguish.
Restoration of non-uniform exposure motion blurred image
Show abstract
Restoring motion-blurred image is the key technologies in the opto-electronic detection system. The imaging sensors such as CCD and infrared imaging sensor, which are mounted on the motion platforms, quickly move together with the platforms of high speed. As a result, the images become blur. The image degradation will cause great trouble for the succeeding jobs such as objects detection, target recognition and tracking. So the motion-blurred images must be restoration before detecting motion targets in the subsequent images. On the demand of the real weapon task, in order to deal with targets in the complex background, this dissertation uses the new theories in the field of image processing and computer vision to research the new technology of motion deblurring and motion detection. The principle content is as follows: 1) When the prior knowledge about degradation function is unknown, the uniform motion blurred images are restored. At first, the blur parameters, including the motion blur extent and direction of PSF(point spread function), are estimated individually in domain of logarithmic frequency. The direction of PSF is calculated by extracting the central light line of the spectrum, and the extent is computed by minimizing the correction between the fourier spectrum of the blurred image and a detecting function. Moreover, in order to remove the strip in the deblurred image, windows technique is employed in the algorithm, which makes the deblurred image clear. 2) According to the principle of infrared image non-uniform exposure, a new restoration model for infrared blurred images is developed. The fitting of infrared image non-uniform exposure curve is performed by experiment data. The blurred images are restored by the fitting curve.