This PDF file contains the front matter associated with SPIE Proceedings Volume 8920, including the Title Page, Copyright information, Table of Contents, Invited Panel Discussion, and Conference Committee listing.

The image restoration algorithms based on time-frequency domain computation is high maturity and applied widely in engineering. To solve the high-speed implementation of these algorithms, the TFDC hardware architecture is proposed. Firstly, the main module is designed, by analyzing the common processing and numerical calculation. Then, to improve the commonality, the iteration control module is planed for iterative algorithms. In addition, to reduce the computational cost and memory requirements, the necessary optimizations are suggested for the time-consuming module, which include two-dimensional FFT/IFFT and the plural calculation. Eventually, the TFDC hardware architecture is adopted for hardware design of real-time image restoration system. The result proves that, the TFDC hardware architecture and its optimizations can be applied to image restoration algorithms based on TFDC, with good algorithm commonality, hardware realizability and high efficiency.

Image matching is a fundamental task in computer vision. It is used to establish correspondence between two images taken at different viewpoint or different time from the same scene. However, its large computational complexity has been a challenge to most embedded systems. This paper proposes a single FPGA-based image matching system, which consists of SIFT feature detection, BRIEF descriptor extraction and BRIEF matching. It optimizes the FPGA architecture for the SIFT feature detection to reduce the FPGA resources utilization. Moreover, we implement BRIEF description and matching on FPGA also. The proposed system can implement image matching at 30fps (frame per second) for 1280x720 images. Its processing speed can meet the demand of most real-life computer vision applications.

In order to implement real-time detection of hedgehopping target in large view-field infrared (LVIR) image, the paper proposes a fast algorithm flow to extract the target region of interest (ROI). The ground building region was rejected quickly and target ROI was segmented roughly through the background classification. Then the background image containing target ROI was matched with previous frame based on a mean removal normalized product correlation (MRNPC) similarity measure function. Finally, the target motion area was extracted by inter-frame difference in time domain. According to the proposed algorithm flow, this paper designs the high-speed real-time signal processing hardware platform based on FPGA + DSP, and also presents a new parallel processing strategy that called function-level and task-level, which could parallel process LVIR image by multi-core and multi-task. Experimental results show that the algorithm can extract low altitude aero target with complex background in large view effectively, and the new design hardware platform could implement real time processing of the IR image with 50000x288 pixels per second in large view-field infrared search system (LVIRSS).

This paper presents a hardware-efficient design for the one-dimensional (1-D) discrete Fourier transform (DFT). Once the 1-D DFT is formulated as the cyclic convolution form, the first-order moments-based structure can be used as the basic computing unit for the DFT computation, which only contains a control module, a statistical module and an accumulation module. The whole calculation process only contains shift operations and additions, with no need for multipliers and large memory. Compared with the traditional DA-based structure for DFT, the proposed design has better performance in terms of the area-throughput ratio and the power consumption, especially when the length of DFT is slightly longer. Similar efficient designs can be obtained for other computations, such as the DCT/IDCT, DST/IDST, digital filter and correlation, by transforming them into the forms of the first-order moments-based cyclic convolution.

This paper proposes an optimized and efficient matching method based on Particle Swarm Optimization (PSO) for image matching. PSO is an efficient intelligent algorithm in image matching. It is a kind of stochastic optimized algorithm developed by Eberhart and Kennedy in 1995. In this paper, the application of PSO is focused on image matching in 3 dimensions with variant angles. The ordinary template matching for the 3 dimensions image matching involves large computational complexity. PSO has been improved in the aspect of self-adaption for convergence. Combining PSO with the individual intelligence, the computation and error rate have been significantly reduced. An extended part of PSO algorithm called multi-swarms is introduced. The multi-swarms PSO (MPSO) is applied to the multi-targets matching in the high dimension space. The performance of MPSO is satisfactory due to the interaction between different swarms such as repulsion and convergence. The Experiments results show that Particle Swarm Optimization Algorithm is much faster in the image matching tasks. MPSO has a good performance in multi-targets matching which involves huge computation complexity.

This paper introduces and evaluates a new algorithm for the computation of type-III discrete Hartley transforms (DHT) of length N = 2ⁿ. The length-N type-III discrete Hartley transforms can be decomposed into several length-16 type-III discrete Hartley transforms based on the radix-2 fast algorithm, and the length-16 type-III discrete Hartley transforms can be computed by first- order moments. It can save a lot of arithmetic operations and the computational complexity of the algorithms is lower than some existing methods. Moreover, this algorithm can be easily implemented.

Differential Evolution (DE) is a simple yet efficient stochastic algorithm for solving real world problems. However, the performance of DE is sensitive to the mutation and crossover strategies and their associated parameters. In this paper, a kind of scale factor generating scheme within the process of search is proposed, named MSFDE, to enhance the performance of DE. In this method, the scale factor is a D dimensional matrix which component is a random number for each difference vector during the iteration. The proposed scheme has been evaluated on a test-suite of 25 benchmark functions provided by CEC 2005 special session on real parameter optimization. The results of the experiments indicate that MDVDE is competitive to classical DE and some other variants on different strategies.

High-quality photorealistic rendering of 3D modeling needs powerful computing systems. On this demand highly efficient management of cluster resources develops fast to exert advantages. This paper is absorbed in the aim of how to improve the efficiency of 3D rendering tasks in cluster. It focuses research on a dynamic feedback load balance (DFLB) algorithm, the work principle of load sharing facility (LSF) and optimization of external scheduler plug-in. The algorithm can be applied into match and allocation phase of a scheduling cycle. Candidate hosts is prepared in sequence in match phase. And the scheduler makes allocation decisions for each job in allocation phase. With the dynamic mechanism, new weight is assigned to each candidate host for rearrangement. The most suitable one will be dispatched for rendering. A new plugin module of this algorithm has been designed and integrated into the internal scheduler. Simulation experiments demonstrate the ability of improved plugin module is superior to the default one for rendering tasks. It can help avoid load imbalance among servers, increase system throughput and improve system utilization.

Now many image super-resolution methods suppose that the optical flows between images should be computed accurately. But really it is very difficult to get them and the models of imaging systems are unknown almost. Thurs perturbation errors always occur in the image super-resolution model. The paper proposes an improved image super-resolution algorithm based on total least squares method. The average image based on images is used as regularized penalty for posteriori probability model. The paper presents the improved Rayleigh quotient format for energy objective function. Then a conjugate gradient algorithm is used to minimize the modified Rayleigh quotient function. The method can minimize two the errors from the sampled low-resolution images and in that perturbation system matrix of high-resolution reconstruction. The test results showed that the algorithm is stable for the perturbation system matrix.

In some cases, there are some special requirements for the vehicle routing problem. Personnel or goods geographically scattered, should be delivered simultaneously to an assigned place by a fleet of vehicles as soon as possible. In this case the objective is to minimize the distance of the longest route among all sub-routes. An improved genetic algorithm was adopted to solve these problems. Each customer has a unique integer identifier and the chromosome is defined as a string of integers. Initial routes are constructed randomly, and then standard proportional selection incorporating elitist is chosen to guarantee the best member survives. New crossover and 2-exchange mutation is adopted to increase the diversity of group. The algorithm was implemented and tested on some instances. The results demonstrate the effectiveness of the method.

A particle-inspired Monte Carlo tree estimation method is proposed to avoid repeating similar simulation and handle the depletion problem in particle filter. Under the inspiration of particles, the method divides the state-space recursively in a top-down manner to form a tree structure that each node in the tree is corresponding to a sub-space. Particles are allocated to the corresponding terminal node during the procedure. Certain size of minimal sub-space or piece is specified to terminate the dividing. Each piece is corresponding to a leaf-node of the tree structure and the prediction probability density in it is approximated by the proportion of its particles in total particles. Instead of importance sampling for each particle, the method takes uniformly random measurements to compute the posterior probability density in each piece. As a result, the method is applied to growth model and has better performance in high SNR environments compared with the Sampling Importance Resampling method.

The autoregressive modeling image interpolation scheme is noticeably closer to ideal interpolation aiming at obtaining a high-resolution (HR) image from its low-resolution (LR) version than conventional methods. The basic idea is to first estimate the covariance of HR image from the covariance of the LR image and then adjust the covariance coefficients of HR image according to a feedback mechanism that takes into account the mutual influence between the estimated missing pixels in a local window. In spite of its impressive performance, the time-consuming computation is usually the bottleneck of the method when it is applied in time-critical scenario. Graphics Processing Units (GPUs) are attractive candidates to expedite the computation process. In this paper, an efficient GPU-based massively parallel version of the autoregressive modeling image interpolation scheme was proposed. Because all pixels which need to be interpolated have no dependence, each estimated pixel is assigned to independent thread in our parallel interpolation scheme. Experimental results show that we reached a speedup of 21.2x when I/O transfer time was taken into account, with respect to the original single-threaded C CPU code with the -O2 compiling optimization.

ImpulseC is based on the C language which can describe highly parallel and multi-process applications. It also generates a underlying hardware description for the dedicated process. To improve the famous bi-cubic interpolation algorithm, we design the bi-cubic convolution template algorithms with better computing performance and higher efficiency. The results of simulation show that the interpolation method not only improves the interpolation accuracy and image quality, but also preferably retains the texture of the image. Based on ImpulseC hardware design tools, we can make use of the compiler features to further parallelize the algorithm so that it is more conducive to the hardware implementation. Based on the Xilinx Spartan3 of XC3S4000 chip, our method achieves the real-time interpolation at the rate of 50fps. The FPGA experimental results show that the stream of output images after interpolation is robust and real-time. The summary shows that the allocation of hardware resources is reasonable. Compared with the existing hand-written HDL code, it has the advantages of parallel speedup. Our method provides a novel idea from C to FPGA-based embedded hardware system for software engineers.

In recent years, Low-rank matrix recovery from corrupted noise matrix has attracted interests as a very effective method in high-dimensional data. And its fast algorithm has become a research focus. This paper we first review the basic theory and typical accelerated algorithms. All these methods are proposed to mitigating the computational burden, such as the iteration count before convergence, especially the frequent large-scale Singular Value Decomposition (SVD). For better convergence, we employ the Augmented Lagrange Multipliers to solve the optimization problem. Recent the endeavors have focused on smaller-scale SVD, especially the method based on submatrix. Finally, we present numerical experiments on large-scale date.

During printing quality inspection, the inspection of color error is an important content. However, the RGB color space is device-dependent, usually RGB color captured from CCD camera must be transformed into CIELAB color space, which is perceptually uniform and device-independent. To cope with the problem, a Markov chain Monte Carlo (MCMC) based algorithms for the RGB to the CIELAB color space transformation is proposed in this paper. Firstly, the modeling color targets and testing color targets is established, respectively used in modeling and performance testing process. Secondly, we derive a Bayesian model for estimation the coefficients of a polynomial, which can be used to describe the relation between RGB and CIELAB color space. Thirdly, a Markov chain is set up base on Gibbs sampling algorithm (one of the MCMC algorithm) to estimate the coefficients of polynomial. Finally, the color difference of testing color targets is computed for evaluating the performance of the proposed method. The experimental results showed that the nonlinear polynomial regression based on MCMC algorithm is effective, whose performance is similar to the least square approach and can accurately model the RGB to the CIELAB color space conversion and guarantee the color error evaluation for printing quality inspection system.

The microvasculature network of retina plays an important role in the study and diagnosis of retinal diseases (age-related macular degeneration and diabetic retinopathy for example). Although it is possible to noninvasively acquire high-resolution retinal images with modern retinal imaging technologies, non-uniform illumination, the low contrast of thin vessels and the background noises all make it difficult for diagnosis. In this paper, we introduce a novel retinal vessel extraction algorithm based on gradient vector flow and matched filtering to segment retinal vessels with different likelihood. Firstly, we use isotropic Gaussian kernel and adaptive histogram equalization to smooth and enhance the retinal images respectively. Secondly, a multi-scale matched filtering method is adopted to extract the retinal vessels. Then, the gradient vector flow algorithm is introduced to locate the edge of the retinal vessels. Finally, we combine the results of matched filtering method and gradient vector flow algorithm to extract the vessels at different likelihood levels. The experiments demonstrate that our algorithm is efficient and the intensities of vessel images exactly represent the likelihood of the vessels.

Images acquired in free breathing using contrast enhanced ultrasound exhibit a periodic motion that needs to be compensated for if a further accurate quantification of the hepatic perfusion analysis is to be executed. In this work, we present an algorithm to compensate the respiratory motion by effectively combining the PCA (Principal Component Analysis) method and block matching method. The respiratory kinetics of the ultrasound hepatic perfusion image sequences was firstly extracted using the PCA method. Then, the optimal phase of the obtained respiratory kinetics was detected after normalizing the motion amplitude and determining the image subsequences of the original image sequences. The image subsequences were registered by the block matching method using cross-correlation as the similarity. Finally, the motion-compensated contrast images can be acquired by using the position mapping and the algorithm was evaluated by comparing the TICs extracted from the original image sequences and compensated image subsequences. Quantitative comparisons demonstrated that the average fitting error estimated of ROIs (region of interest) was reduced from 10.9278 ± 6.2756 to 5.1644 ± 3.3431 after compensating.

Due to the curve of the coronary artery and the overlap, cross between its branches, some of its information is lost in the 3D-2D imaging process, which may leads to the inaccuracy in reconstructing three-dimensional vascular tree structure from angiographic images. In this paper, a new three-dimensional reconstruction method using overlap detection for 3-D projection is proposed to improve this problem, and experiments proves that the method can raise the accuracy of the reconstruction.

Functional magnetic resonance imaging (fMRI) is an advanced non-invasive data acquisition technique to investigate the neural activity in human brain. In addition to localize the functional brain regions that is activated by specific cognitive task, fMRI can also be utilized to measure the task-related functional interactions among the active regions of interest (ROI) in the brain. Among the variety of analysis tools proposed for modeling the connectivity of brain regions, Granger causality analysis (GCA) measure the directions of information interactions by looking for the lagged effect among the brain regions. In this study, we use fMRI and Granger Causality analysis to investigate the effective connectivity of brain network induced by viewing several kinds of expressional faces. We focus on four kinds of facial expression stimuli: fearful, angry, happy and neutral faces. Five face selective regions of interest are localized and the effective connectivity within these regions is measured for the expressional faces. Our result based on 8 subjects showed that there is significant effective connectivity from STS to amygdala, from amygdala to OFA, aFFA and pFFA, from STS to aFFA and from pFFA to aFFA. This result suggested that there is an information flow from the STS to the amygdala when perusing expressional faces. This emotional expressional information flow that is conveyed by STS and amygdala, flow back to the face selective regions in occipital-temporal lobes, which constructed a emotional face processing network.

Digital pathological image retrieval plays an important role in computer-aided diagnosis for breast cancer. The retrieval results of an unknown pathological image, which are generally previous cases with diagnostic information, can provide doctors with assistance and reference. In this paper, we develop a novel pathological image retrieval method for breast cancer, which is based on stain component and probabilistic latent semantic analysis (pLSA) model. Specifically, the method firstly utilizes color deconvolution to gain the representation of different stain components for cell nuclei and cytoplasm, and then block Gabor features are conducted on cell nuclei, which is used to construct the codebook. Furthermore, the connection between the words of the codebook and the latent topics among images are modeled by pLSA. Therefore, each image can be represented by the topics and also the high-level semantic concepts of image can be described. Experiments on the pathological image database for breast cancer demonstrate the effectiveness of our method.

Image mosaic has widely applications value in the fields of medical image analysis, and it is a technology that carries on the spatial matching to a series of image which are overlapped with each other, and finally builds a seamless and high quality image which has high resolution and big eyeshot. In this paper, the method of grayscale cutting pseudo-color enhancement was firstly used to complete the mapping transformation from gray to the pseudo-color, and to extract SIFT features from the images. And then by making use of a similar measure of NCC (normalized cross correlation - Normalized cross-correlation), the method of RANSAC (Random Sample Consensus) was used to exclude the pseudofeature points right in order to complete the exact match of feature points. Finally, seamless mosaic and color fusion were completed by using wavelet multi-decomposition. The experiment shows that the method we used can effectively improve the precision and automation of the medical image mosaic, and provide an effective technical approach for automatic medical image mosaic.

With the development of computer aided navigation system, more and more tissues shall be reconstructed to provide more useful information for surgical pathway planning. In this study, we aimed to propose a registration framework for different reconstructed tissues from multi-modalities based on some fiducial points on lateral ventricles. A male patient with brain lesion was admitted and his brain scans were performed by different modalities. Then, the different brain tissues were segmented in different modality with relevant suitable algorithms. Marching cubes were calculated for three dimensional reconstructions, and then the rendered tissues were imported to a common coordinate system for registration. Four pairs of fiducial markers were selected to calculate the rotation and translation matrix using least-square measure method. The registration results were satisfied in a glioblastoma surgery planning as it provides the spatial relationship between tumors and surrounding fibers as well as vessels. Hence, our framework is of potential value for clinicians to plan surgery.

Carotid atherosclerosis is the major cause of ischemic stroke, a leading cause of mortality and disability. Morphology and structure features of carotid plaques are the keys to identify plaques and monitoring the disease. Manually segmentation on the ultrasonic images to get the best-fitted actual size of the carotid plaques based on physicians personal experience, namely "gold standard", is a important step in the study of plaque size. However, it is difficult to qualitatively measure the segmentation error caused by the operator's subjective factors. In order to reduce the subjective factors, and the uncertainty factors of quantification, the experiments in this paper were carried out. In this study, we firstly designed a carotid artery phantom, and then use three different beam-forming algorithms of medical ultrasound to simulate the phantom. Finally obtained plaques areas were analyzed through manual segmentation on simulation images. We could (1) directly evaluate the different beam-forming algorithms for the ultrasound imaging simulation on the effect of carotid artery; (2) also analyze the sensitivity of detection on different size of plaques; (3) indirectly reflect the accuracy of the manual segmentation base on segmentation results the evaluation.

The paper proposed a segmentation method combining both local and global threshold techniques to efficiently segment the cell images. Firstly, the image would be divided into several parts, and the Otsu operation would be used to each of them to detect details. Secondly, main body of the objects would be filtered out by a global threshold algorithm. Finally, based on the previous steps, more advanced segmentation outcomes can be achieved. The experimental results show that this algorithm made better performance at detail recognition, such as the cell antennas, which should be very helpful and important in the medical area.

As an important optical molecular imaging technique, bioluminescence tomography (BLT) offers an inexpensive and sensitive means for non-invasively imaging a variety of physiological and pathological activities at cellular and molecular levels in living small animals. The key problem of BLT is to recover the distribution of the internal bioluminescence sources from limited measurements on the surface. Considering the sparsity of the light source distribution, we directly formulate the inverse problem of BLT into an l₀-norm minimization model and present a smoothed l₀-norm (SL0) based reconstruction algorithm. By approximating the discontinuous l0 norm with a suitable continuous function, the SL0 norm method solves the problem of intractable computational load of the minimal l₀ search as well as high sensitivity of l₀-norm to noise. Numerical experiments on a mouse atlas demonstrate that the proposed SL0 norm based reconstruction method can obtain whole domain reconstruction without any a priori knowledge of the source permissible region, yielding almost the same reconstruction results to those of l₁ norm methods.