This PDF file contains the front matter associated with SPIE Proceedings Volume 7251, including the Title Page, Copyright information, Table of Contents, Introduction (if any), and the Conference Committee listing

This paper describes two separate visual-based inspection procedures used at CANDU nuclear power generating stations. The techniques are quantitative in nature and are delivered and operated in highly radioactive environments with access that is restrictive, and in one case is submerged. Visual-based inspections at stations are typically qualitative in nature. For example a video system will be used to search for a missing component, inspect for a broken fixture, or locate areas of excessive corrosion in a pipe. In contrast, the methods described here are used to measure characteristic component dimensions that in one case ensure ongoing safe operation of the reactor and in the other support reactor refurbishment. CANDU reactors are Pressurized Heavy Water Reactors (PHWR). The reactor vessel is a horizontal cylindrical low-pressure calandria tank approximately 6 m in diameter and length, containing heavy water as a neutron moderator. Inside the calandria, 380 horizontal fuel channels (FC) are supported at each end by integral end-shields. Each FC holds 12 fuel bundles. The heavy water primary heat transport water flows through the FC pressure tube, removing the heat from the fuel bundles and delivering it to the steam generator. The general design of the reactor governs both the type of measurements that are required and the methods to perform the measurements. The first inspection procedure is a method to remotely measure the gap between FC and other in-core horizontal components. The technique involves delivering vertically a module with a high-radiation-resistant camera and lighting into the core of a shutdown but fuelled reactor. The measurement is done using a line-of-sight technique between the components. Compensation for image perspective and viewing elevation to the measurement is required. The second inspection procedure measures flaws within the reactor's end shield FC calandria tube rolled joint area. The FC calandria tube (the outer shell of the FC) is sealed by rolling its ends into the rolled joint area. During reactor refurbishment, the original FC calandria tubes are removed, potentially scratching the rolled joint area and, thereby, compromising the seal with the new FC calandria tube. The procedure involves delivering an inspection module having a radiation-resistant camera, standard lighting, and a structured lighting projector. The surface is inspected by rotating the module within the rolled joint area. If a flaw is detected, its depth and width are gauged from the profile variation of the structured lighting in a captured image. As well, the diameter profile of the area is measured from the analysis of a series of captured circumferential images of the structured lighting profiles on the surface.

We have been developing a rapid proto-typing display system which can verify an appearance of final product in finishing and painting industry. In this system, it is necessary to measure detail information of hand position and shape to recognize the worker's instruction. Therefore, we apply a rapid hand measurement which combine a roughly detecting of hand position and shape by spatial encoding method with IR projection. For detecting of hand position, non-linearity interval strips are used for detecting objects that are lower than constant height. The interval of strips is devised in relation to an angle of camera axis to make equal the height in detecting. For detecting of hand shape, the temporal and spatial encoding pattern is projected only an area of hand position. This measurement is enough rough because our prototyping display system need only to classify the shape of tracing, touching, pushing, and picking. Therefore, the limited process with limited area is possible to reconstruct the shape of hand very fast. A practical result shows that the position and shape recognition is performed about one second; and operator comment that such the time delay doesn't become a stress as for actual hand operation.

We enhance an existing in-circuit, inline tester for printed circuit assemblies (PCA) by video-based automatic optical inspection (Video-AOI). Our definition of video is that we continuously capture images of a moving PCA, such that each PCA component is contained in multiple images, taken under varying viewing conditions like angle, time, camera settings or lighting. This can then be exploited for an efficient detection of faults. The first part of our paper focuses on the parameters of such a Video-AOI system and shows how they can be determined. In the second part, we introduce techniques to capture and preprocess a video of a PCA, so that it can be used for inspection.

Automatic vehicle Make and Model Recognition (MMR) systems provide useful performance enhancements to vehicle recognitions systems that are solely based on Automatic Number Plate Recognition (ANPR) systems. Several vehicle MMR systems have been proposed in literature. In parallel to this, the usefulness of multi-resolution based feature analysis techniques leading to efficient object classification algorithms have received close attention from the research community. To this effect, Contourlet transforms that can provide an efficient directional multi-resolution image representation has recently been introduced. Already an attempt has been made in literature to use Curvelet/Contourlet transforms in vehicle MMR. In this paper we propose a novel localized feature detection method in Contourlet transform domain that is capable of increasing the classification rates up to 4%, as compared to the previously proposed Contourlet based vehicle MMR approach in which the features are non-localized and thus results in sub-optimal classification. Further we show that the proposed algorithm can achieve the increased classification accuracy of 96% at significantly lower computational complexity due to the use of Two Dimensional Linear Discriminant Analysis (2DLDA) for dimensionality reduction by preserving the features with high between-class variance and low inter-class variance.

The feed manufacturers can control the composition of feed in relation to their feed value. But, in practice, an important issue is still pending: the poultries can reject a batch of feed with optimal nutritional characteristics. This rejection is often accompanied by undesirable and incomprehensible reactions (e.g. pecks in multiple directions) leading to negative consequences for the animal as well as the poultry breeder and the firm. Zootechnical studies are dealing with two main research areas: modeling the poultry feeding behavior and linking it with the poultry perception, especially vision. Currently, a study is undertaken to define the poultry feeding behavior and to point out feeds corresponding to different reactions. As for the perception, visual aspects of feed seem to be involved. While the objective of the study is to make it possible to control the visual quality of feed according to animal behavior, the goal of the present work is to discriminate between feeds of different firms based on visual features extracted from feed images. This discrimination by visual features could be linked with the poultry feeding behaviour and be an effective foundation for the control of the feed acceptability by visual aspects. In this paper, we assess the relevance of color and texture features and we show how these features are involved in the discrimination process between feed images.

During industrial forging of big hot metallic shells, it is necessary to regularly measure the dimensions of the parts, especially the inner and outer diameters and the thickness of the walls, in order to decide when to stop the forging process. The inner and outer diameters of the shells range from 4 to 6 meters and to measure them a large ruler is placed horizontally at the end of the shell. Two blacksmiths standing on each side of the ruler at about ten meters from it visually reads the graduations on the ruler in order to determine the inner and outer diameters from which the thickness of the wall is determined. This operation is carried out several times during a forging process and it is very risky for the blacksmiths due to the high temperature of the shell when the measurement is done. Also, it is error prone and the result is rather inaccurate. In order to improve the working conditions, for the safety of the blacksmiths, and for a faster and more accurate measurement, a system based on two commercially available Time Of Flight (TOF) laser scanners for the measurement of cylindrical shell diameters during the forging process has been developed. The advantages of using laser scanners are that they can be placed very far from the hot shell, more than 15 meters, while at the same time giving an accurate point cloud from which 3D views of the shell can be reconstructed and diameter measurements done. Moreover, better dimensional measurement accuracy is achieved in less time with the laser system than with the conventional method using a large ruler. The system has been successfully used to measure the diameter of cold and hot cylindrical metallic shells.

Stain release is the degree to which a stained substrate approaches its original unsoiled appearance as a result of care procedure. Stain release has a significant impact on the pricing of the fabric and, hence, needs to be quantified in an objective manner. In this paper, an automatic approach for the objective assessment of fabric stain release that utilizes region-based statistical snakes, is presented. This deformable contour approach employs a pressure energy term in the parametric snake model in conjunction with statistical information (hence, statistical snakes) extracted from the image to segment the stain and subsequently assign a stain release grade. This algorithm has been parallelized on a General Purpose Graphical Processing Unit (GPGPU) for accelerated and simultaneous segmentation of multiple stains on a fabric. The computational power of the GPGPU is attributed to its hardware and software architecture, which enables multiple and identical snake kernels to be processed in parallel on several streaming processors. The detection and segmentation results of this machine vision scheme are illustrated as part of the validation study. These results establish the efficacy of the proposed approach in producing accurate results in a repeatable manner. In addition, this paper presents a comparison between the benchmarking results for the algorithm on the CPU and the GPGPU.

In this paper, we present a novel method for fingerprint verification. A unique characteristic of our method is the use of direction images and local features in the matching process. A direction center is computed from the direction image and used as a reference point for aligning fingerprints. Fingerprint matching is performed in two stages. In the first stage, we compute the correlation between the direction images of the two fingerprints. In the second stage, we compare various features derived from fingerprint minutiae. The first stage acts as a filtering procedure that rejects fingerprints based on the global directional patterns of the ridges. The second stage verifies the local characteristics of the fingerprint minutiae. The two-stage matching process results in a robust procedure that minimizes verification errors.

In the context of fine structure extraction, lots of methods have been introduced, and, particularly in pavement crack detection. We can distinguish approaches based on a threshold, employing mathematical morphology tools or neuron networks and, more recently, techniques with transformations, like wavelet decomposition. The goal of this paper is to introduce a 2D matched filter in order to define an adapted mother wavelet and, then, to use the result of this multi-scale detection into a Markov Random Field (MRF) process to segment fine structures of the image. Four major contributions are introduced. First, the crack signal is replaced by a more real one based on a Gaussian function which best represents the crack. Second, in order to be more realistic, i.e. to have a good representation of the crack signal, we use a 2D definition of the matched filter based on a 2D texture auto-correlation and a 2D crack signal. The third and fourth improvements concern the Markov network designed in order to allow cracks to be a set of connected segments with different size and position. For this part, the number of configurations of sites and potential functions of the MRF model are completed.

Detection of small vessels is a challenging task for navy, coast guard and port authority for security purposes. Vessel identification is more complex as compared to other object detection because of its variability in shapes, features and orientations. Current methods for vessel detection are primarily based on segmentation techniques which are not as efficient and also require different algorithms for visible and infrared images. In this paper, a new vessel detection technique is proposed employing anomaly detection. The input intensity image is first converted to feature space using difference of Gaussian filters. Then a detector filter in the form of Mahalanobis distance is applied to the feature points to detect anomalies whose characteristics are different from their surroundings. Anomalies are detected as bright spots in both visible and infrared image. The larger the gray value of the pixels the more anomalous they are to be. The detector output is then post-processed and a binary image is constructed where the boat edges with strong variance relative to the background are identified along with few outliers from the background. The resultant image is then clustered to identify the location of the vessel. The main contribution in this paper is developing an algorithm which can reliably detect small vessels in visible and infrared images. The proposed method is investigated using real-life vessel images and found to perform excellent in both visible and infrared images with the same system parameters.

Images of high geometrical complexity are found in various applications in the fields of image processing and computer vision. Medical imaging is such an application, where the processing of digitized images reveals vital information for the therapeutic or diagnostic algorithms. However, the segmentation of these images has been proved to be one of the most challenging topics in modern computer vision algorithms. The light interaction with tissues and the geometrical complexity with the tangent objects are among the most common reasons that many segmentation techniques nowadays are strictly related to specific applications and image acquisition protocols. In this paper a sophisticated segmentation algorithm is introduced that succeeds into overcoming the application dependent accuracy levels. This algorithm is based on morphological sequential filtering, combined with a watershed transformation. The results on various biomedical test images present increased accuracy, which is independent of the image acquisition protocol. This method can provide researchers with a valuable tool, which makes the classification or the follow-up faster, more accurate and objective.

Images of biological objects in transmission electron microscopy (TEM) are particularly noisy and low contrasted, making their processing a challenging task to accomplish. During these last years, several software tools were conceived for the automatic or semi-automatic acquisition of TEM images. However, tools for the automatic analysis of these images are still rare. Our study concerns in particular the automatic identification of artificial membranes at medium magnification for the control of an electron microscope. We recently proposed a segmentation strategy in order to detect the regions of interest. In this paper, we introduce a complementary technique to improve contour recognition by a statistical validation algorithm. Our technique explores the profile transition between two objects. A transition is validated if there exists a gradient orthogonal to the contour that is statistically significant.

Bundle adjustment is a minimization method frequently used to refine the structure and motion parameters of a moving camera. In this work, we present a Newton-based approach to enhance the accuracy of the estimated motion parameters in the bundle adjustment framework. The key issue is to first parameterize the motion variables of a camera on the manifold of the Euclidean motion by using the underlying Lie group structure of the motion representation. Second, it is necessary to formulate the bundle adjustment cost function and derive the corresponding gradient and the Hessian formulation on the manifold using the concepts of differential geometry. This results in a more compact derivation of the Hessian which allows us to use its complete form in the minimization process. Compared to the Levenberg-Marquardt scheme, the proposed algorithm is shown to provide more accurate results while having a comparable complexity although the latter uses an approximate form of the Hessian. The experimental results we performed on simulated and real image sets are evidence that demonstrate our claims.

The ability of a robot to localise itself and simultaneously build a map of its environment (Simultaneous Localisation and Mapping or SLAM) is a fundamental characteristic required for autonomous operation of the robot. Vision Sensors are very attractive for application in SLAM because of their rich sensory output and cost effectiveness. Different issues are involved in the problem of vision based SLAM and many different approaches exist in order to solve these issues. This paper gives a classification of state-of-the-art vision based SLAM techniques in terms of (i) imaging systems used for performing SLAM which include single cameras, stereo pairs, multiple camera rigs and catadioptric sensors, (ii) features extracted from the environment in order to perform SLAM which include point features and line/edge features, (iii) initialisation of landmarks which can either be delayed or undelayed, (iv) SLAM techniques used which include Extended Kalman Filtering, Particle Filtering, biologically inspired techniques like RatSLAM, and other techniques like Local Bundle Adjustment, and (v) use of wheel odometry information. The paper also presents the implementation and analysis of stereo pair based EKF SLAM for synthetic data. Results prove the technique to work successfully in the presence of considerable amounts of sensor noise. We believe that state of the art presented in the paper can serve as a basis for future research in the area of vision based SLAM. It will permit further research in the area to be carried out in an efficient and application specific way.

This paper discusses an edge-direction-based template matching algorithm that allows to detect industrial objects despite perspective distortion. We construct a deformable template by decomposing a shape model into independent parts, where the deformation is restricted to, e.g., a homography. The deformable template in combination with a coarse-to-fine strategy allows to overcome the speed limitations of an exhaustive template matching of a 3D search range. The relevant size of the model that is used for the search at the highest pyramid level is not reduced. Therefore, we do not suffer the speed limitations that prior methods have. Furthermore, enforcing a consistent polarity in each part, but ignoring different polarities between different parts allows us to efficiently and robustly detect untextured metallic objects that are encountered in typical factory automation scenarios. Finally, we present results of an experimental evaluation with respect to speed, robustness and accuracy.

To treat a color image in a holistic manner, we take a unique approach to color image processing. In previous work we have proposed a new feature image of which pixel holds the number of frequency of color vectors. This feature image that we call Frequency Image is made from a special color histogram of an image and presents a distribution of frequency of color. In this paper, first, we review the basic idea of Frequency Image and present a new analysis of this image. Next, we explain the effective applications such as color edge detection, color uniformity inspection, focusing and local exposure compensation. Then we propose a new approach to color image segmentation and demonstrate some experimental results. Finally, we discuss some issues and advantages of using the Frequency Image. A Frequency Image is useful to reduce the dimension of an original color image and to arrange a classification by the frequency of color vectors. Therefore we can utilize this image effectively in various color image-processing applications.

Active Thermography has become a powerful tool in the field of non-destructive testing (NDT) in recent years. This infrared thermal imaging technique is used for non-contact inspection of materials and components by visualizing thermal surface contrasts after a thermal excitation. The imaging modality combined with the possibility of detecting and characterizing flaws as well as determining material properties makes Active Thermography a fast and robust testing method even in industrial-/production environments. Nevertheless, depending on the kind of defect (thermal properties, size, depth) and sample material (CFRP carbon fiber reinforced plastics, metal, glass fiber) or sample structure (honeycomb, composite layers, foam), active thermography can sometimes produce equivocal results or completely fails in certain test situations. The aim of this paper is to present examples of results of Active Thermography methods conducted on aircraft components compared to various other (imaging) NDT techniques, namely digital shearography, industrial x-ray imaging and 3D-computed tomography. In particular we focus on detection limits of thermal methods compared to the above-mentioned NDT methods with regard to: porosity characterization in CFRP, detection of delamination, detection of inclusions and characterization of glass fiber distributions.

Focusing on 3D object recognition for handling-robot tasks, we developed a registration method for point data measured from a real object and model surfaces. On the basis of the iterative-closest-point (ICP) algorithm, we proposed a registration technique that deforms model shapes instead of correcting measured range data including distance errors. We call our technique a "viewpoint-dependent remodeling ICP" algorithm. Even when a laser range finder only is used, this technique can reduce the effects of errors depending on surface characteristics such as colors and reflectance properties. In the preliminary stages, the relationships between distance errors and surface characteristics of points on object surfaces are determined and added to the models. In object recognition stages, we measure point data, and do registration while changing the model position and attitude and deforming the model shape. The deformation depends on the relationships and the relative positions of the model surfaces and the sensor position. In preliminary experimental tests, we measured distances to black and white papers and evaluated the distance errors. Moreover, we simulated recognizing the bottle covered with these papers. In this simulation, it was verified that our technique has convergence and improves accuracy of correspondence estimations between measured data and models.

This work outlines a system in which a stereo camera may effectively track a user's face and hands in three dimensions. Given this information, a method for controlling objects in three dimensions is also described. The system begins by finding faces. If more than one face is found in the image, the algorithm uses depth information to isolate the face that is closest to the camera. The algorithm then gathers information about the user's skin tone by examining the content of the face found. For much of the processing, only the hue and saturation components are used after applying an HSV to RGB transformation given the camera output. The skin tone information in tandem with depth is then used to isolate the user's hands, and track them in three dimensions. To be used as an effective interface, the system uses information of the two hands relative to the user's face. In controlling an object in three dimensions, if the user would like to move the object up, he or she simply positions both hands above his or her face. Similar commands allow the user to apply a translational factor in three dimensions, as well as applying yaw and roll when wanted.

As of December 2008, the two Mars rover spacecraft Spirit and Opportunity have collected more than 4 years worth of data from nine imaging instruments producing greater than 200k images which includes both raw image data from spacecraft instruments and images generated by post-processing algorithms developed by NASA's Multimission Image Processing Laboratory (MIPL). This paper describes a prototype software system that allows scientists to browse and data-mine the images produced from NASA's Mars Exploratory Rover (MER) missions with emphasis on the automatic detection of images containing rocks that are of interest for geological research. We highlight two aspects of our prototype system: (1) software design for mining remote data repositories, (2) a computationally efficient image search engine for detecting MER images that containing rocks. Datatype abstractions made at the software design level allow users to access and visualize the source data through a single simple-to-use interface when the underlying data may originate from a local or remote image repository. Data mining queries into the MER image data are specified over chronological intervals denoted (sols) as each interval is a solar day. As in other mining applications, an automatic detection and classification algorithm is used to compute a relevance score that represents how relevant a given recorded image is to the user-specified query. Query results are presented as list of records, sorted by their relevance score, which the user may then visualize and investigate to extract information of interest. Several standard image analysis tools are provided for investigation of 2D images (e.g., histogram equalization, edge detection, etc.) and, when available, stereoscopic data is integrated with the image data using multiple windows which show both the 2D image and 3D surface geometry. The combination of data mining and a high-quality visualization interface provides MER researchers unprecedented access to the recorded data.

In this paper, we develop rectangular Gaussian kernels, i.e. all the rotated versions of the first order partial derivatives of the 2D nonsymmetrical Gaussian functions, which are used to convolve with the test images for edge extraction. By using rectangular kernels, one can have greater flexibility to smooth high frequency noise while keeping the high frequency edge details. When using larger kernels for edge detection, one can smooth more high frequency noise at the expense of edge details. Rectangular kernels allow us to smooth more noise along one direction and detect better edge details along the other direction, which improve the overall edge detection results especially when detecting line pattern edges. Here we propose two new approaches in using rectangular Gaussian kernels, namely the pattern-matching method and the quadratic method. The magnitude and directional edge from these two methods are computed based on the convolution results of the small neighborhood of the edge point with the rectangular Gaussian kernels along different directions.

Remote sensing is widely used assess the destruction from natural disasters and to plan relief and recovery operations. How to automatically extract useful features and segment interesting objects from digital images, including remote sensing imagery, becomes a critical task for image understanding. Unfortunately, current research on automated feature extraction is ignorant of contextual information. As a result, the fidelity of populating attributes corresponding to interesting features and objects cannot be satisfied. In this paper, we present an exploration on meaningful object extraction integrating reflecting surfaces. Detection of specular reflecting surfaces can be useful in target identification and then can be applied to environmental monitoring, disaster prediction and analysis, military, and counter-terrorism. Our method is based on a statistical model to capture the statistical properties of specular reflecting surfaces. And then the reflecting surfaces are detected through cluster analysis.

This paper presents a simulation of automatic 3D acquisition and post-processing pipeline. The proposed methodology is applied to a LASER triangulation based scanner and a 6 degrees of freedom (DOF) robotic arm simulation. The viewpoints are computed by solving a set covering problem to reduce the number of potential positions. The quality of the view plan is determined by its length and the percentage of area of the object's surface it covers. Results are presented and discussed on various shapes. The article also presents future work concerning the implementation of the proposed method on a real system.

Drip paintings by the American Abstract Expressionist Jackson Pollock have been analyzed through computer image methods, generally in support of authentication studies. The earliest and most thoroughly explored methods are based on an estimate of a "fractal dimension" by means of box-counting algorithms, in which the painting's image is divided into ever finer grids of boxes and the proportion of boxes containing some paint is counted. The plot of this proportion (on a log-log scale) reveals scaling or fractal properties of the work. These methods have been extended in a number of ways, including multifractal analysis, where an information measure replaces simple box paint occupancy. Recent studies suggest that it is unlikely that any single measure, including those based on such box counting, will yield highly accurate authentication; for example, a broad class of highly artificial angular sketches created in software reveal the same "fractal" properties as genuine Pollock paintings. Others have argued that this result precludes the value of such fractal-based features for such authentication. We show theoretically that even if a visual feature (taken alone) is "uninformative," such a feature can enhance discrimination when it is combined in a classifier with other features-even if these other features are themselves also individually uninformative. We describe simple classifiers for distinguishing genuine Pollocks from fakes based on multiple features such as fractal dimension, topological genus, "energy" in oriented spatial filters, and so forth. We trained linear-discriminant and nearest-neighbor classifiers using these features and found that our classifiers gave slightly improved recognition accuracy on human generated drip paintings. Most importantly, we found that although fractal features, taken alone might have low discriminative power, such features improved accuracy in multi-feature classifiers. We conclude that it is premature to reject the use of visual features based on box-counting statistics for the authentication of Pollock's dripped works, particularly if such measures are used in conjunction with multiple features, machine learning and art material studies and connoisseurship.

A recent theory claims that some Renaissance artists, as early as 1425, secretly traced optically projected images during the execution of some passages in some of their works, nearly a quarter millennium before historians of art and of optics have secure evidence anyone recorded an image this way. Key evidence adduced by the theory's proponents includes the trelliswork in the right panel of Robert Campin's Merode altarpiece triptych (c. 1425-28). If their claim were verified for this work, such a discovery would be extremely important to the history of art and of image making more generally: the Altarpiece would be the earliest surviving image believed to record the projected image of an illuminated object, the first step towards photography, over 400 years later. The projection theory proponents point to teeny "kinks" in the depicted slats of one orientation in the Altarpiece as evidence that Campin refocussed a projector twice and traced images of physically straight slats in his studio. However, the proponents rotated the digital images of each slat individually, rather than the full trelliswork as a whole, and thereby disrupted the relative alignment between the images of the kinks and thus confounded their analysis. We found that when properly rotated, the kinks line up nearly perfectly and are consistent with Campin using a subtly kinked straightedge repeatedly, once for each of the slats. Moreover, the proponents did not report any analysis of the other set of slats-the ones nearly perpendicular to the first set. These perpendicular slats are straight across the break line of the first set-an unlikely scenario in the optical explanation. Finally, whereas it would have been difficult for Campin to draw the middle portions of the slats perfectly straight by tracing a projected image, it would have been trivially simple had he used a straightedge. Our results and the lack of any contemporaneous documentary evidence for the projection technique imply that Campin used a simple mechanical aid-such as a minutely kinked straightedge or a mahl stick commonly used in the early Renaissance-rather than a very complex optical projector and procedure, undocumented from that time.

Automatic sheet inspection in banknote production has been used as a standard quality control tool for more than a decade. As more and more print techniques and new security features are established, total quality in bank note printing must be guaranteed. This aspect has a direct impact on the research and development for bank note inspection systems in general in the sense of technological sustainability. It is accepted, that print defects are generated not only by printing parameter changes, but also by mechanical machine parameter changes, which will change unnoticed in production. Therefore, a new concept for a multi-sensory adaptive learning and classification model based on Fuzzy-Pattern- Classifiers for data inspection and machine conditioning is proposed. A general aim is to improve the known inspection techniques and propose an inspection methodology that can ensure a comprehensive quality control of the printed substrates processed by printing presses, especially printing presses which are designed to process substrates used in the course of the production of banknotes, security documents and others. Therefore, the research and development work in this area necessitates a change in concept for banknote inspection in general. In this paper a new generation of FPGA (Field Programmable Gate Array) based real time inspection technology is presented, which allows not only colour inspection on banknote sheets, but has also the implementation flexibility for various inspection algorithms for security features, such as window threads, embedded threads, OVDs, watermarks, screen printing etc., and multi-sensory data processing. A variety of algorithms is described in the paper, which are designed for and implemented on FPGAs. The focus is based on algorithmic approaches.

We present two novel Poisson noise Maximum Likelihood based methods for identifying the individual returns within mixed pixels for Amplitude Modulated Continuous Wave rangers. These methods use the convolutional relationship between signal returns and the recorded data to determine the number, range and intensity of returns within a pixel. One method relies on a continuous piecewise truncated-triangle model for the beat waveform and the other on linear interpolation between translated versions of a sampled waveform. In the single return case both methods provide an improvement in ranging precision over standard Fourier transform based methods and a decrease in overall error in almost every case. We find that it is possible to discriminate between two light sources within a pixel, but local minima and scattered light have a significant impact on ranging precision. Discrimination of two returns requires the ability to take samples at less than 90 phase shifts.

For the 9000 train accidents reported each year in the European Union [1], the Recording Strip (RS) and Filling-Card (FC) related to the train activities represent the only usable evidence for SNCF (the French railway operator) and most of National authorities. More precisely, the RS contains information about the train journey, speed and related Driving Events (DE) such as emergency brakes, while the FC gives details on the departure/arrival stations. In this context, a complete checking for 100% of the RS was recently voted by French law enforcement authorities (instead of the 5% currently performed), which raised the question of an automated and efficient inspection of this huge amount of recordings. To do so, we propose a machine vision prototype, constituted with cassettes receiving RS and FC to be digitized. Then, a video analysis module firstly determines the type of RS among eight possible types; time/speed curves are secondly extracted to estimate the covered distance, speed and stops, while associated DE are finally detected using convolution process. A detailed evaluation on 15 RS (8000 kilometers and 7000 DE) shows very good results (100% of good detections for the type of band, only 0.28% of non detections for the DE). An exhaustive evaluation on a panel of about 100 RS constitutes the perspectives of the work.

In this paper we present a new fuzzy logic based approach for automatic optimized features allocation. The technique is used for improved automatic alignment and classification of silicon wafers and chips that are used in the electronic industry. The proposed automatic image processing approach was realized and experimentally demonstrated in real industrial application with typical wafers. The automatic features allocation and grading supported the industrial requirements and could replace human expert based inspection that currently is performed manually.

This paper describes a novel embedded system capable of estimating 3D positions of surfaces viewed by a stereoscopic rig consisting of a pair of calibrated cameras. Novel theoretical and technical aspects of the system are tied to two aspects of the design that deviate from typical stereoscopic reconstruction systems: (1) incorporation of an 10x zoom lens (Rainbow- H10x8.5) and (2) implementation of the system on an embedded system. The system components include a DSP running μClinux, an embedded version of the Linux operating system, and an FPGA. The DSP orchestrates data flow within the system and performs complex computational tasks and the FPGA provides an interface to the system devices which consist of a CMOS camera pair and a pair of servo motors which rotate (pan) each camera. Calibration of the camera pair is accomplished using a collection of stereo images that view a common chess board calibration pattern for a set of pre-defined zoom positions. Calibration settings for an arbitrary zoom setting are estimated by interpolation of the camera parameters. A low-computational cost method for dense stereo matching is used to compute depth disparities for the stereo image pairs. Surface reconstruction is accomplished by classical triangulation of the matched points from the depth disparities. This article includes our methods and results for the following problems: (1) automatic computation of the focus and exposure settings for the lens and camera sensor, (2) calibration of the system for various zoom settings and (3) stereo reconstruction results for several free form objects.

The purpose of this study is to classify the types of coconut plantation. To this end, we compare several classifiers such as Maximum Likelihood, Minimum Distance, Parallelepiped, Mahalanobis and Support Vector Machines (SVM). The contribution of textural informations and spectral informations increases the separability of different classes and then increases the performance of classification algorithms. Before comparing these algorithms, the optimal windows size, on which the textural information are computed, as well as the SVM parameters are first estimated. Following this study, we conclude that SVM gives very satisfactory results for coconut field type mapping.

We propose a new color correction approach which, as opposed to existing methods, take advantages of a given pair of two color face images (probe and gallery) in the color face recognition (FR) framework. In the proposed color correction method, the color-flow vector and color-flow eigenspace model are developed to generate color corrected probe images. The main contribution of this paper is threefold: 1) the proposed method can reliably compensate the non-linear photic variations imposed on probe face images comparing to traditional color correction techniques; 2) to the best of our knowledge, for the first time, we conduct extensive experiment studies to compare the effectiveness of various color correction methods to deal with photometrical distortions in probe images; 3) the proposed method can significantly enhance the recognition performance degraded by severely illuminant probe face images. Two standard face databases CMU PIE and XM2VTSDB were used to demonstrate the effectiveness of the proposed color correction method. The usefulness of the proposed method in the color FR is shown in terms of both absolute and comparative recognition performances against four traditional color correction solutions of White balance, Gray-world, Retinex, and Color-by-correlation.

Building footprint extraction from GIS imagery/data has been shown to be extremely useful in various urban planning and modeling applications. Unfortunately, existing methods for creating these footprints are often highly manual and rely largely on architectural blueprints or skilled modelers. Although there has been quite a lot of research in this area, most of the resultant algorithms either remain unsuccessful or still require human intervention, thus making them infeasible for practical large-scale image processing systems. In this work, we present novel LiDAR and aerial image processing and fusion algorithms to achieve fully automated and highly accurate extraction of building footprint. The proposed algorithm starts with initial building footprint extraction from LiDAR point cloud based on an iterative morphological filtering approach. This initial segmentation result, while indicating locations of buildings with a reasonable accuracy, may however produce inaccurate building footprints due to the low resolution of the LiDAR data. As a refinement process, we fuse LiDAR data and the corresponding color aerial imagery to enhance the accuracy of building footprints. This is achieved by first generating a combined gradient surface and then applying the watershed algorithm initialized by the LiDAR segmentation to find ridge lines on the surface. The proposed algorithms for automated building footprint extraction have been implemented and tested using ten overlapping LiDAR and aerial image datasets, in which more than 300 buildings of various sizes and shape exist. The experimental results confirm the efficiency and effectiveness of our fully automated building footprint extraction algorithm.

The objective of our work is to develop a tool for automatic analysis of 2D membrane protein crystal images in Transmission Electron Microscopy (TEM). The success of crystallization experiments is evaluated at high magnification. The crystalline structure of a membrane can be observed when no other membranes are superposed. It is therefore necessary to identify mono-layer membranes. In this paper we introduce an algorithm that determines the stacking-level of membranes. Our method determines a quantum, a gray-level quantity that is characteristic of a non-stacked membrane. In this way we are able to label each region qualitatively and construct a stacking-level map that distinguishes from non-stacked to up to four-level stacked membranes. This map provides the regions that will trigger a new image acquisition at higher magnification.

Finger vein authentication is a personal identification technology using finger vein images acquired by infrared imaging. It is one of the newest technologies in biometrics. Its main advantage over other biometrics is the low risk of forgery or theft, due to the fact that finger veins are not normally visible to others. Extracting finger vein patterns from infrared images is the most difficult part in finger vein authentication. Uneven illumination, varying tissues and bones, and changes in the physical conditions and the blood flow make the thickness and brightness of the same vein different in each acquisition. Accordingly, extracting finger veins at their accurate positions regardless of their thickness and brightness is necessary for accurate personal identification. For this purpose, we propose a new finger vein extraction method which is composed of gradient normalization, principal curvature calculation, and binarization. As local brightness variation has little effect on the curvature and as gradient normalization makes the curvature fairly uniform at vein pixels, our method effectively extracts finger vein patterns regardless of the vein thickness or brightness. In our experiment, the proposed method showed notable improvement as compared with the existing methods.

In many production systems, the products are inspected by human operators who observe faults with their naked eye while most of the other manufacturing activities are automated. However, manual inspection is slow and yields subjective results. To defeat this problem, image processing based visual control systems have been integrated to the production systems. The visual system performance depends on the robustness of the image processing techniques. Especially, the thresholding technique plays crucial role if you are inspecting scratches on the products. Since utilizing the constant threshold fails in many cases, we have proposed an adaptive thresholding technique based visual inspection system to detect production faults rapidly and efficiently without hampering the manufacturing process. The proposed visual system also includes rotation invariant properties, which is important to get high speed processing.

The finger vein image acquired with an acquisition system should be properly aligned to proceed with comparing algorithm. However it is not easy to find control the points since the images are naturally blurred with an inherent scattering property. To overcome this problem, we propose a novel finger vein registration method utilizing skin surface information (i.e. wrinkles and outlines). We assumed that finger crooking was insignificant. Images were sampled with intended translation and rotation. Each time, two images were acquired successively by switching the light source; one with infrared light and the other with white light. Degree of rotation and translation of sampled image were calculated using outline features in the white light image and then the infrared image was transformed according to the calculated data. To validate our method, correlation values were computed between identical subjects and different subjects. High correlation values were shown between identical subjects whereas low values were shown between different subjects.