Search

The averaging of multiple pairwise relative motions in a sequence provides a fast and accurate method of camera motion estimation with a wide range of applications, including view registration, robotic path estimation, super-resolution. Since this approach involves averaging in the Lie-algebra of the underlying motion representation, it is non-robust and susceptible to contamination due to outliers in the individual relative motions. In this paper, we introduce a graph-based sampling scheme that efficiently remove such motion outliers. The resulting global motion solution is robust and also provides an empirical estimate of the inherent statistical uncertainty. Example results are provided to demonstrate the efficacy of our approach to incorporating robustness in motion averaging. [ABSTRACT FROM AUTHOR]

In the past decades, most object recognition systems were based on passive approaches. But in the last few years a lot of research was done in the field of active object recognition. In this context, there are several unique problems to be solved, such as the fusion of views and the selection of an optimal next viewpoint. In this paper we present an approach to solve the problem of choosing optimal views (viewpoint selection) and the fusion of these for an optimal 3D object recognition (viewpoint fusion). We formally define the selection of additional views as an optimization problem and we show how to use reinforcement learning for viewpoint training and selection in continuous state spaces without user interaction. In this context we focus on the modeling of the reinforcement learning reward. We also present an approach for the fusion of multiple views based on density propagation, and discuss the advantages and disadvantages of two approaches for the practical evaluation of these densities, namely Parzen estimation and density trees. [ABSTRACT FROM AUTHOR]

Some issues on direct-projected augmented reality (DirectAR) are addressed: the projection may be geometrically distorted due to the non-planar surface (geometric distortion); the projection cannot be seen to user as intended because the position of the projector is not the same as that of the user's viewpoint (viewpoint-ignorant projection); the projection may be modulated by surface color (radiometric distortion); the projected area may not have uniform brightness when the projection is obliquely headed for the surface (uneven projection). We propose an integrated framework for handling all the problems. Experimental results demonstrate that the problems unavoidable in surface-independent DirectAR can be successfully resolved. [ABSTRACT FROM AUTHOR]

In this paper, we describe a novel uncalibrated active stereo system using coded structured light. Structured-light-based active stereo systems generally consist of a camera and projector that require precise precalibration. Therefore, if we can eliminate the precalibration process from the system, the user can merely place the equipment arbitrarily and directly begin scanning the objects. This will greatly improve both the convenience and practicality of the system. In order to achieve this, we propose an original self-calibration method that can be considered as a camera-to-camera self-calibration method in which one of the cameras is replaced with a projector. We also propose a simultaneous 3D reconstruction method that utilizes multiple captured stereo pairs to increase the accuracy of the 3D estimation. Further, we suggest a simple solution to eliminate the ambiguity of scaling by attaching a laser pointer to the projector, which is important for the practical use of the 3D reconstruction. [ABSTRACT FROM AUTHOR]

We present a method for the reconstruction of a specular surface, using a single camera viewpoint and the reflection of a planar target placed at two different positions. Contrarily to most specular surface reconstruction algorithms, our method makes no assumption on the regularity or continuity of the specular surface, and outputs a set of 3D points along with corresponding surface normals, all independent from one another. A point on the specular surface can be reconstructed if its corresponding pixel in the image has been matched to its source in both of the target planes. We present original solutions to the problem of dense point matching and planar target pose estimation, along with reconstruction results in real-world scenarii. [ABSTRACT FROM AUTHOR]

Despite the long history of studies on weak-perspective projection and the intuitive notion that this model has better numerical stability compared to the perspective model under appropriate conditions, there lacks a deep understanding about the error characteristics of motion and depth recovery process under this model. In this paper, we present the differential approach to SFM under weak-perspective projection. Based on this approach, error behavior which governs the formation of motion ambiguities under weak-perspective is investigated. Recovery of depth information and its distortion are also discussed. Regions where ordinal depth can be extracted is shown to be following a simple relationship under all types of motion, a different result compared to the case of perspective projection. [ABSTRACT FROM AUTHOR]

This paper proposes a new method based on self-calibration to estimate the ball's 3D position in broadcast soccer video. According to the physical limitation, the ball's 3D position is estimated through the camera position and the ball's virtual shadow, which is the point of intersection between the playfield and the line through the camera's optical center and the ball. First, the virtual shadow is computed by the homography between playfield and image plane. For the image having enough corresponding points, the map is determined directly; for those images not having enough these points, their homographies are estimated through global motion estimation. Then, based on self-calibrating for rotating and zooming camera, and the homography, the camera's position in the playfield is estimated. Experiments show that the proposed method can extract ball's 3D position information without referring to other object with assuming height and obtain promising results. [ABSTRACT FROM AUTHOR]

We present a stereo vision based global self-localization strategy for tiny autonomous mobile robots in a well-known dynamic environment. Global localization is required for an initial startup or when the robot loses track of its pose during navigation. Existing approaches are based on dense range scans, active beacon systems, artificial landmarks, bearing measurements using omni-directional cameras or bearing/range calculation using single frontal cameras, while we propose feature based stereo vision system for range calculation. Location of the robot is estimated using range measurements with respect to distinct landmarks such as color transitions, corners, junctions and line intersections. Unlike methods based on angle measurement, this method requires only two distinct landmarks. Simulation results show that robots can successfully localize themselves whenever two distinct landmarks are observed. As such marked minimization of landmarks for vision based self-localization of robots has been achieved. [ABSTRACT FROM AUTHOR]

We present an algorithm that synchronizes two short video sequences where an object undergoes ballistic motion against stationary scene points. The object's motion and epipolar geometry are exploited to guide the algorithm to the correct synchronization in an iterative manner. Our algorithm accurately synchronizes videos recorded at different frame rates, and takes few iterations to converge to sub-frame accuracy. We use synthetic data to analyze our algorithm's accuracy under the influence of noise. We demonstrate that it accurately synchronizes real video sequences, and evaluate its performance against manual synchronization. [ABSTRACT FROM AUTHOR]

Moving object detection is a challenging task for night security because of bad video quality. In this paper, we propose a robust real time objects detection method for night visual surveillance based on human visual system. By measuring contrast information variation in multiple successive frames, a spatio-temporal contrast change image (CCI) is formed. Then the multi-frame correspondence technology is employed to robustly extract salient motions or moving objects from CCI. Since CCI is a statistical measurement of variation based on human visual system, the proposed method is effective at night and better than traditional detection methods. Experiments on real scene show that the method based on contrast feature is effective for night object detection and tracking, our approach is also robust to camera scale variation as well as low computation cost. [ABSTRACT FROM AUTHOR]

Recently, it has been shown that invariants on motions can be extracted from sequential images and these can be applied for recognizing dynamic events from images viewed from arbitrary viewpoints. These invariants are called space-time invariants since they are defined in space and time. Unfortunately, the existing space-time invariants are limited for planar motions viewed from affine cameras. In this paper, we propose a method for computing space-time invariants on general 3D motions viewed from projective cameras. Furthermore, we show that by using the epipolar geometry derived from the mutual projection of cameras, the stability of space-time invariants can be improved drastically. The extracted invariants are applied for distinguishing non-rigid 3D motions from video sequences viewed from arbitrary viewpoints. [ABSTRACT FROM AUTHOR]

This paper presents a motion capture system using two cameras that is capable of estimating a constrained set of human postures in real time. We first obtain a 3D shape model of a person to be tracked and create a posture dictionary consisting of many posture examples. The posture is estimated by hierarchically matching silhouettes generated by projecting the 3D shape model deformed to have the dictionary poses onto the image plane with the observed silhouette in the current image. Based on this method, we have developed a virtual fashion show system that renders a computer graphics-model moving synchronously to a real fashion model, but wearing different clothes. [ABSTRACT FROM AUTHOR]

We propose a hierarchical method for 3D rigid motion estimation between two 3D data sets of objects represented by triangular meshes. Multiresolution surfaces are generated from the original surface of each object. These surfaces are decomposed into small patches based on estimated geodesic distance and curvature information. In our method, segment-to-segment matching to recover rigid motions at each resolution level of surfaces is performed. Motion results from low resolution surface matching are propagated to higher resolution surface matching in order to generate a spatial constraint for similar segment selection. Our approach can recover 3D rigid motion of both rigid body and nonrigid body (with partial rigid areas). The method was tested to estimate rigid motions of 3D data obtained by the Cyberware scanner. [ABSTRACT FROM AUTHOR]

This paper presents a robust player tracking method for sports video analysis. In order to track agile player stably and robustly, we employ multiple models method, with a mean shift procedure corresponding to each model for player localization. Furthermore, we define pseudo measurement via fusing the measurements obtained by mean shift procedure. And the fusing coefficients are built from two likelihood functions: one is image based likelihood; the other is motion based association probability. Experimental results show effectiveness of our method in the hard case of player tracking literature. [ABSTRACT FROM AUTHOR]

In this paper we propose an effective and robust approach for detecting corner points on a given binary image. Unlike other corner detection methods the proposed method is non-parametric in nature, that is, it does not require any input parameter. The proposed method is based on mathematical morphology. It makes use of morphological skeleton for detecting corner points. Convex corner points are obtained by intersecting the morphological boundary and the corresponding skeleton, where as the concave corner points are obtained by intersecting the boundary and the skeleton of the complement image. Experimental results show that the proposed method is more robust and efficient in detecting corner points. Keywords: Corner points, Morphological skeleton, Non-parametric. [ABSTRACT FROM AUTHOR]

In this paper, we present a new method to deal with specular highlights in correspondence search. The proposed method is essentially based on the specular-free two-band image that we introduce to deal with specular reflection. For given input images, specular-free two-band images are generated using simple pixel-wise computations in real-time. Specular-free two-band images are then used to compute per-pixel raw matching costs. By using the specular-free two-band images instead of input images, reliable raw matching costs that are independent of the specularities of image pixels are obtained. As a result, we can find correct correspondences even in the presence of specular highlights. Experimental results show that the proposed method successfully produces accurate disparity maps for stereo images with specular highlights. [ABSTRACT FROM AUTHOR]

To recognize or identify objects it is desirable to use features which are minimally affected by changes in lighting and non-stationary noise. This requires accurate estimation of both signal and noise. In response to this challenge, this paper proposes a method for estimation of non-stationary isotropic noise based on steering filters to directions perpendicular and parallel to the local signal. From the filter responses in this direction equations for signal and noise are obtained which lead to an edge detection method dependent solely upon local signal-to-noise ratio. The proposed method is compared to various common edge detection methods from the literature, on synthetic and real images. Quantitative improvement is demonstrated on synthetic images and qualitative improvement on real images. [ABSTRACT FROM AUTHOR]

In this paper, we design and implement a novel method for constructing a mixed triangle/quadrangle mesh from the 3D space curves (rims) estimated from the profiles of an object in an image sequence without knowing the original 3D topology of the object. To this aim, a contour data structure for representing visual hull, which is different from that for CT/MRI, is introduced. In this paper, we (1) solve the "branching structure" problem by introducing some additional "directed edge", and (2) extract a triangle/quadrangle closed mesh from the contour structure with an algorithm based on dynamic programming. Both theoretical demonstration and real world results show that our proposed method has sufficient robustness with respect to the complex topology of the object, and the extracted mesh is of high quality. [ABSTRACT FROM AUTHOR]

Many active safety technologies for the driver support system are developed. Most of the traffic accidents are caused by driver's inattentive or drowsy. We are developing a driver support system that protects from traffic accidents by these causes. Our purpose is to detect the driver's face region. A lot of face detection methods are proposed, but there is not a technique addressing every environment inside the car. In this paper, we propose a skin detection method by the unique reflection characteristics of the materials. We developed the skin detection system, and confirmed the effectiveness by the evaluation experiment. [ABSTRACT FROM AUTHOR]

A fast block matching algorithm, namely Fast Walsh Search, is proposed for motion estimation in block-based video coding. In our approach, target blocks in current frame and their candidates in reference frame are projected onto Walsh Hadamard domain, allowing early rejection of mismatch candidates to reduce computation requirement. Moreover, we introduce a new method called block pyramid matching that re-uses many previous calculations to further lessen the computation load of our approach. Experimental results show that the proposed algorithm can achieve more accurate motion estimation than the popular three-step-search and diamond search with slight increase in computation requirement only. [ABSTRACT FROM AUTHOR]

Inkjet and laser printers, and photocopiers are being increasingly used in criminal activities such as counterfeiting currency, creating forged documents, illegitimate business transactions and terrorism related acts. Identification of the printing process greatly aids in detecting such activities and is extensively used in the field of document examination. In this paper, we propose the use of image processing techniques in identifying the printing process used to generate a document. The characteristics of the various types of non-impact printing methods used by photocopiers, inkjet and laser printers are studied using colour image processing. HSV color space and, in particular, hue images at high-resolution, distribution of isolated spots in the vicinity of hue edge pixels and periodicities in edge intensity profiles distinguish between the different printing processes. Our initial study indicates their promise in replicating the results traditionally obtained by document examiners using a microscope or through chemical analysis. [ABSTRACT FROM AUTHOR]

In this paper we mimic a biological visual strategy to extract salient contours from complex scenes. Psychophysical and physiological studies show that the response to the stimulus within the receptive field is affected by the presence of surrounding stimuli— the response is suppressed significantly by similarly oriented stimuli in the surround while this suppression is converted to strong facilitation with the addition of collinear stimuli in the surround. According to this property of visual perception, we enhance salient contours and at the same time reduce the interference of the extraneous elements. Our results show the feasibility of the proposed method. [ABSTRACT FROM AUTHOR]

In this paper, we describe image feature as parameterized model and formulate feature detection as robust model fitting problem. It can detect global feature easily without parameter transformation, which is needed by Hough Transform methods. We adopt RANSAC paradigm to solve the problem. It is immune to outliers and can deal with image contains multiple features and noisy pixels. In the voting stage of RANSAC, in contrast with previous methods which need distance computation and comparison, we apply Bresenham algorithm to generate pixels in the inlier region of the feature and use the foreground pixels in this region to vote the potential feature. It greatly improves the efficiency and can detect spatially-linked features easily. Experimental results with both synthetic and real images are reported. [ABSTRACT FROM AUTHOR]

Eye feature extraction is of crucial importance for face recognition. Deformable template is an efficient model for this task. However, it usually suffers from the problem of local minima. To avoid local minima, in this paper, a new energy minimization process is proposed, which emphasizes on local properties of energy terms. The minimization process is divided into three steps. The iris is located firstly. Then the eye boundaries are adjusted. Finally, all energy terms are activated to tune the eye template. Each step needs not to be split to some sub-steps. Empirical comparison with other minimization processes shows the superiority of the proposed process in terms of both efficiency and accuracy. [ABSTRACT FROM AUTHOR]

The problem of detecting local image features that are invariant to scale, orientation, illumination and viewpoint changes is a critical issue in many computer vision applications. The challenges involve localizing the image features accurately in the spatial and frequency domains and describing them with a stable analytical representation. In this paper we address these two issues by proposing a new non-linear scale-space implementation that improves the localization accuracy of the SIFT [3] local features. Furthermore we propose a simple adjustment to the standard SIFT descriptor and show that the modified version is more robust to affine changes. [ABSTRACT FROM AUTHOR]

Image classification could be treated as an effective solution to enable keyword-based semantic image retrieval, while feature selection is a key issue in categorization. In this paper, we propose a novel strategy by using feature selection in learning semantic concepts of image categories. To choose representative and informative features for an image category and meanwhile reduce noisy features, a feature selection strategy is proposed. In the feature selection stage, salient patches are first detected by SIFT descriptor and clustered by DENCLUE algorithm. Then the pointwise mutual information between the salient patches and the image category is calculated to evaluate the important patches and construct the visual vocabulary for the category. Based on the selected visual features, the SVM classifier is applied to categorization. The experimental results on Corel image database demonstrate that the proposed feature selection approach is very effective in image classification and visual concept learning. [ABSTRACT FROM AUTHOR]

A five layered, event driven hierarchical framework for generic sports video classification has been proposed in this paper. The top layer classifications are based on a few popular audio and video content analysis techniques like short-time energy and Zero Crossing Rate (ZCR) for audio and Hidden Markov Model (HMM) based techniques for video, using color and motion as features. The lower layer classifications are done by applying game specific rules to recognize major events of the game. The proposed framework has been successfully tested with cricket and football video sequences. The event-related classifications bring us a step closer to the ultimate goal of semantic classifications that would be ideally required for sports highlight generation. [ABSTRACT FROM AUTHOR]

In this paper, we address one very important industrial application of computer vision - automatic classification of materials. In our work, we have considered materials that are mixtures of two or more elements. Such materials are called alloys. It is observed at the microscopic level that an alloy is composed of small randomly distributed crystals of varying shapes and sizes called grains. Also, the color and hence the intensity of the grains vary in alloys. Generally, this shape-size-intensity distribution of the grains is different for different materials. This means micrographs obtained from different materials form texture-like images that differ from one material to another in appearance. Therefore, in principle, any texture analysis method may be used for material classification. In our method, we propose to extract textural features corresponding to grain geometry and intensity and use them for analysis and classification of alloys. These features are extracted via gray-scale morphological operations and are measured in terms of Size-Intensity-Diagram (SID) and Tri-variate Pattern Spectrum (TPS) coefficients. In our experiments, we achieved 83.43% and 89.43% classification accuracies in cases of SID and TPS, respectively. This demonstrates the effectiveness of the proposed method for material classification which in turn confirms that our choice of features is indeed appropriate for the purpose. [ABSTRACT FROM AUTHOR]

We propose a new method for classification of photometric factors, such as diffuse reflection, specular reflection, attached shadow, and cast shadow. For analyzing real images, we utilize the photometric linearization method which was originally proposed for image synthesis. First, we show that each pixel can be photometrically classified by the simple comparison of the pixel intensity. Our classification algorithm requires neither 3D shape information nor color information of the scene. Then, we show that the accuracy of the photometric linearization can be improved by introducing a new classification-based criterion to the linearization process. Experimental results show that photometric factors can be correctly classified without any special device. [ABSTRACT FROM AUTHOR]

To efficiently classify and track video objects in a surveillance application, it is essential to reduce the amount of streaming data. One solution is to segment the video into background, i.e. stationary objects, and foreground, i.e. moving objects, and then discard the background. One such motion segmentation algorithm that has proven reliable is the Stauffer and Grimson algorithm. This paper investigates how different color spaces affect the segmentation result in terms of noise and shadow sensitivity. Shadows are especially problematic since they not only distort shape but can also result in falsely connected objects that will complicate tracking and classification. Therefore, a new decision kernel for the segmentation algorithm is presented. This kernel alters the probability of foreground detection to reduce shadows and to increase the chance of correct segmentation for objects with a skin tone color, e.g. faces. [ABSTRACT FROM AUTHOR]

We developed an automatic object segmentation system to separate the foreground objects from the background scene in the MPEG video sequence. The system consists of two modules: the background modeling and updating module and the foreground object extraction module. For the first module and comparing to existing methods, the background model can be constructed no matter whether there exist moving foreground objects or not. In addition, the background model is capable of handling the illumination changes and intrusive but motionless targets by using the short-term approach and long-term approach, respectively, to keep updating the background model. For the second module, the noises and shadows are eliminated and the holes are filled in order to reduce the false and the missing foreground detection components, respectively. Furthermore, one particular function in this module is the automatic digital matting, which can be applied to have visually accurate segmentation result for the foreground objects. [ABSTRACT FROM AUTHOR]

We propose an efficient way to account for spatial smoothness in foreground-background segmentation of video sequences. Most statistical background modeling techniques regard the pixels in an image as independent and disregard the fundamental concept of smoothness. In contrast, we model smoothness of the foreground and background with a Markov random field, in such a way that it can be globally optimized at video frame rate. As a background model, the mixture-of-Gaussian (MOG) model is adopted and enhanced with several improvements developed for other background models. Experimental results show that the MOG model is still competitive, and that segmentation with the smoothness prior outperforms other methods. [ABSTRACT FROM AUTHOR]

We propose an empirical computational model for generating an interpretation of a video shot based on our proposed principle of perceptual prominence. The principle of perceptual prominence captures the key aspects of mise-en-scène required for interpreting a video scene. We present a novel approach for applying perceptual grouping principles to the spatio-temporal domain of video. Our spatio-temporal perceptual grouping scheme, applied on blob tracks, makes use of a specified spatio-temporal coherence model. A high level semantic interpretation of scenes is done using the mise-en-scène features and the perceptual prominence computed for the perceptual clusters. [ABSTRACT FROM AUTHOR]

In this paper, we present a new scalable 3D object representation and learning method to recognize many objects. Scalability is one of the important issues in object recognition to reduce memory and recognition time. The key idea of scalable representation is to combine a feature sharing concept with view clustering in part-based object representation (especially a CFCM: common frame constellation model). In this representation scheme, we also propose a fully automatic learning method: appearance-based automatic feature clustering and sequential construction of view-tuned CFCMs from labeled multi-views and multi-objects. We applied this learning scheme to 40 objects with 216 training views. Experimental results show the scalable learning results in almost constant recognition performance relative to the number of objects. [ABSTRACT FROM AUTHOR]

This paper presents a practical method for hypothesizing hand locations and subsequently recognizing a discrete number of poses in image sequences. In a typical setting the user is gesturing in front of a single camera and interactively performing gesture input with one hand. The approach is to identify likely hand locations in the image based on discriminative features of colour and motion. A set of exemplar templates is stored in memory and a nearest neighbour classifier is then used for hypothesis verification and pose estimation. The performance of the method is demonstrated on a number of example sequences, including recognition of static hand gestures and a navigation by pointing application. [ABSTRACT FROM AUTHOR]

A methodology for determining the level of confidence of a sub-region in the overall classification of a given face image affected due to varying expressions, illuminations and partial occlusions is presented in this paper. The technique for obtaining the weights for each individual region of the test image is based on a measure of optical flow between that test image and a face model. Individual image regions or the modules are also assigned additional weights by arranging them in the order of their importance in classification. The approach presented is applicable mainly in scenarios where the number of samples in the training set is too little. A K-nearest neighbor distance measure is used in classifying each module of the test image after dimensionality reduction. A total score is calculated for each training class based on the classification result of each module and its associated weights. Considerable increase in recognition accuracy has been observed for PCA, LDA and ICA based linear subspace approaches when implemented using the proposed technique. [ABSTRACT FROM AUTHOR]

The face pattern is described by pairs of template-based histogram and Fisher projection orientation under the paradigm of AdaBoost learning in this paper. We assume that a set of templates are available first. To avoid making strong assumptions about distributional structure while still retaining good properties for estimation, the classical statistical model, histogram, is used to summarize the response of each template. By introducing a novel "integral histogram image", we can compute histograms rapidly. Then we turn to Fisher linear discriminant for each template to project histograms from d-dimensional to one-dimensional subspace. Best features, used to describe face pattern, are selected by AdaBoost learning. The results of preliminary experiments demonstrate that the selected features are much more powerful to represent the face pattern than the simple rectangle features used by Viola and Jones and some variants. [ABSTRACT FROM AUTHOR]

We developed a system including two modules: the texture analysis module and the texture synthesis module. The analysis module is capable of analyzing an input image and performing the training process by using this image data. According to the training non-periodic or periodic pattern, we use different sampling methods to have different amount of patches in order to reduce the emergences of the seams of the output synthesized image. In addition, the properties of principal component analysis (PCA) are used to reduce the dimensions of the data representation and to recombine the appearance of the features (i.e. eigenvectors). Then the vector quantization (VQ) algorithm is employed to reduce the time spent on matching comparison. For the synthesis module, the training data is used to synthesize a large output texture, or is employed to replace the removed regions of an image. The multi-resolution approach is applied to accelerate the procedure of our algorithm: the down-sampling step is the training process and the up-sampling step is in the order of reconstructing (or synthesizing) the large removed region without needing to assign initial random values or approximate values. Therefore, our system can rapidly obtain a high image quality and promising result. [ABSTRACT FROM AUTHOR]

Many activities may be characterized by a sequence of key frames that are related to important changes in motion rather than dominant characteristics that persist over a long sequence of frames. To detect such changes, we define a transformation operator at every time instant, which relates the past to the future states. One of the useful quantities associated with numerical range of an operator is the eigenvalue. In the literature, eigenvalue-based approaches have been studied extensively for many modeling tasks. These rely on gross properties of the data and are not suitable to detect subtle changes. We propose an antieigenvalue - based measure to detect key frames. Antieigenvalues depend critically on the turning of the operator, whereas eigenvalues represent the amount of dilation along the eigenvector directions aligned with the direction of maximum variance. We demonstrate its application to activity modeling and recognition using two datasets: a motion capture dataset and the UCF human action dataset. [ABSTRACT FROM AUTHOR]

In this paper, we try to use graphical model based probabilistic inference methods to solve the problem of contour matching, which is a fundamental problem in computer vision. Specifically, belief propagation is used to develop the contour matching framework. First, an undirected loopy graph is constructed by treating each point of source contour as a graphical node. Then, the distances between the source contour points and the target contour points are used as the observation data, and supplied to this graphical model. During message transmission, we explicitly penalize two kinds of incorrect correspondences: many-to-one correspondence and cross correspondence. A final geometrical mapping is obtained by minimizing the energy function and maximizing a posterior for each node. Comparable experimental results show that better correspondences can be achieved. [ABSTRACT FROM AUTHOR]

OK-Quantization Theory for the digitization in value ensures the reconstructivity of the probabilistic density function of the image. This paper shows some experimental demonstrations to reduce the number of the gray levels, and shows mainly that there is a necessary analytical relationship between sampling and quantization based on the equivalence relationship between two kinds of the integral, Riemann and Lebesgue integrals for calculating the volume of the image. Experimental demonstrations are also shown in this paper. [ABSTRACT FROM AUTHOR]

Detection and tracking of moving objects is very important in various ways. Concerning the detection of moving objects by stationary cameras, the background looks different as the illumination changes. In this paper, we consider a particular image in an image sequence as the sum of a reference image containing the background and a difference image containing the moving objects but not the background. We show that a reference image and difference images can be obtained as the independent components of input images by Independent Component Analysis. Moving objects can then be located on the reference image and the difference images. Experimental results show that the proposed approach produces accurate detection of moving objects even if illumination changes. [ABSTRACT FROM AUTHOR]

We propose a new method to recover 3D structures of artificial objects from scene pictures using orthogonality and parallelism. A new transformation group, "semi-metric space," is defined to describe the scenes of artificial objects consisting of orthogonal and parallel line features. A metric invariant called conic dual to the circular points has a simple diagonal form in the semi-metric space. Furthermore, under some assumptions, the metric reconstruction is possible using some affine properties. The algorithms are verified with real images captured with a camera in a commercial mobile phone. [ABSTRACT FROM AUTHOR]

When space points and camera optical center lie on a twisted cubic, no matter how many corresponding pairs there are from space points to their image points, camera projection matrix cannot be uniquely determined, in other words, the configuration of camera and space points in this case is critical for camera parameter estimation. In practice, it is important to detect this critical configuration before the estimated camera parameters are used. In this work, a new method is introduced to detect this critical configuration, which is based on an effective criterion function constructed from an invariant relationship between six space points and their corresponding image points. The advantage of this method is that no explicit computation on camera projection matrix or optical center is needed. Simulations show it is quite robust and stable against noise. Experiments on real data show the criterion function can be faithfully trusted for camera parameter estimation. [ABSTRACT FROM AUTHOR]

In this paper, we analyze multiple view geometry under projections from 4D space to 3D space and show that it can represent multiple view geometry under the projection of space with time. In particular, we show that multifocal tensors defined under space-time projections can be computed from non-rigid object motions viewed from multiple cameras with arbitrary translational motions. We also show that they are very useful for generating images of non-rigid object motions viewed from cameras which have arbitrary translational motions. The method is implemented and tested in real and synthetic images. [ABSTRACT FROM AUTHOR]

Real-time estimation of a camera's pose relative to an object is still an open problem. The difficulty stems from the need for fast and robust detection of known objects in the scene given their 3D models, or a set of 2D images or both. This paper proposes a method that conducts a statistical analysis of the appearance of model patches from all possible viewpoints in the scene and incorporates the 3D geometry during both matching and the pose estimation processes. Thereby the appearance information from the 3D model and real images are combined with synthesized images in order to learn the variations in the multiple view feature descriptors using PCA. Furthermore, by analyzing the computed visibility distribution of each patch from different viewpoints, a reliability measure for each patch is estimated. This reliability measure is used to further constrain the classification problem. This results in a more scalable representation reducing the effect of the complexity of the 3D model on the run-time matching performance. Moreover, as required in many real-time applications this approach can yield a reliability measure for the estimated pose. Experimental results show how the pose of complex objects can be estimated efficiently from a single test image. [ABSTRACT FROM AUTHOR]

In this paper, we present an efficient algorithm for reconstructing 3D head model from a single 2D image based on using a 3D eigenhead model. This system is composed of two components, offline training of the eigenhead model and online reconstruction of a 3D head model. For the first part, we propose a new 3D head alignment algorithm based on an iterative coarse-to-fine scheme to establish dense point correspondences between 3D head model in the cylindrical coordinate to align the 3D head models in the training data set. In addition, we apply the radial basis function technique to establish dense correspondences between each 3D face model and a reference face model, followed by the principal component analysis technique to compute the statistical eigenhead model. For the 3D face reconstruction from a single image, the proposed algorithm finds the best linear combination of the eigenhead bases that minimizes an energy function composed of distances between the corresponding facial feature points and a one-way partial Haussdorf distance between the facial contours in the image domain. This energy minimization is accomplished by the iterative Levenberg-Marquardt algorithm with the initial guess determined by solving a linear system derived from the image projection constraints for the corresponding facial feature points. Experimental results show that the proposed 3D face reconstruction algorithm provides satisfactory results and takes less than 10 seconds on a regular PC. [ABSTRACT FROM AUTHOR]

Level set based segmentation has been used with and without shape priors, to approach difficult segmentation problems in several application areas. This paper addresses two limitations of the classical level set based segmentation approaches: They usually deliver just two regions - one foreground and one background region, and if some prior information is available, they are able to take into account just one prior but not more. In these cases, one object of interest is reconstructed but other possible objects of interest and unfamiliar image structures are suppressed. The approach we propose in this paper can simultaneously handle an arbitrary number of regions and competing shape priors. Adding to that, it allows the integration of numerous pose invariant shape priors, while segmenting both known and unknown objects. Unfamiliar image structures are considered as separate regions. We use a region splitting to obtain the number of regions and the initialization of the required level set functions. In a second step, the energy of these level set functions is robustly minimized and similar regions are merged in a last step. All these steps are considering given shape priors. Experimental results demonstrate the method for arbitrary numbers of regions and competing shape priors. [ABSTRACT FROM AUTHOR]

This paper presents a unified variational framework for seamlessly integrating prior segmentation information into non-rigid registration procedures. Under this framework, in addition to the forces arise from the similarity measure in seeking for detailed correspondence, another set of forces generated by the prior segmentation contours can provide an extra guidance in assisting the alignment process towards a more meaningful, stable and noise-tolerant procedure. Local correlation (LC) is being used as the underlying similarity measures to handle intensity variations. We present several 2D/3D examples on synthetic and real data. [ABSTRACT FROM AUTHOR]

This paper addresses the problem of super resolution - obtaining a single high-resolution image given a set of low resolution images which are related by small displacements. We employ a reconstruction based approach using MRF-MAP formalism, and use approximate optimization using graph cuts to carry out the reconstruction. We also use the same formalism to investigate high resolution expansions from single images by deconvolution assuming that the point spread function is known. We present a method for the estimation of the point spread function for a given camera. Our results demonstrate that it is possible to obtain super-resolution preserving high frequency details well beyond the predicted limits of magnification. [ABSTRACT FROM AUTHOR]