It has been over a decade since the "Eigenfaces" approach to automatic face recognition, and other appearance-based methods, made an impression on the computer vision research community and helped spur interest in vision systems being used to support biometrics and human-computer interface. In this paper I give a personal view of the original motivation for the work, some of the strengths and limitation of the approach, and progress in the years since. Appearance-based approaches to recognition complement feature- or shape-based approaches, and a practical face recognition system should have elements of both. Eigenfaces is not a general approach to recognition, but rather one tool out of many to be applied and evaluated in the appropriate context.

The paper proposes a vision-based system for adaptively inferring the interactional intention of a person coming close to a robot, which plays an important role in the succeeding stage of human/robot cooperative handling of works/tools in production lines. Here, interactional intention is ranged in the meaning of the intention to interact/operate with the robot, which is proposed to be estimated by the human head moving path during an incipient period of time. To implement this intention inference capability, first, human entrance is detected and is modeled by an ellipse to supply information about the head position. Second, B-spline technique is used to approximate the trajectory with reduced control points in order that the system acquires information about the human motion direction and the curvature of the motion trajectory. Finally, Hidden Markov Models (HMMs) are applied as the adaptive inference engines at the stage of inferring the human interactional intention. The HMM algorithm with a stochastic pattern matching capability is extended to supply whether or not a person has an intention toward the robot at the incipient time. The reestimation process here models the motion behavior of an human worker when he has or doesn't have the intention to operate the robot. Experimental results demonstrate the adaptability of the inference system using the extended HMM algorithm for filtering out motion deviation over the trajectory.

In this paper, we address the problem of recovering the camera radial distortion coefficients from one image. The approach that we propose uses a special kind of snakes called radial distortion snakes. Radial distortion snakes behave like conventional deformable contours, except that their behavior are globally connected via a consistent model of image radial distortion. Experiments show that radial distortion snakes are more robust and accurate than conventional snakes and manual point selection.

This paper presents a new approach to shape and motion estimation based on geometric primitives and relations in a model-based framework. A description of a scene in terms of structured geometric elements sharing relationships allows to derive a parametric model with Euclidian constraints, and a camera model is also proposed to reduce the problem dimensionality. It leads to a sequential MAP estimation, that gives accurate and comprehensible results on real images.

We consider the problem of reconstructing architectural scenes from multiple photographs taken from arbitrary viewpoints. The original contribution is the use of a map as a source of geometric constraints to obtain in a fast and simple way a detailed model of a scene. We suppose that images are uncalibrated and have at least one planar structure as a faade for exploiting the planar homography induced between world plane and image to calculate a first estimation of the projection matrix. Estimations are improved by using correspondences between images and map. We show how these simple constraints can be used to calibrate the cameras and recover the projection matrices for each viewpoint. Finally, triangulation is used to recover 3D models of the scene and to visualise new viewpoints. Our approach needs minimal a priori information about the camera being used. A working system has been designed and implemented to allow the user to interactively build a model from uncalibrated images from arbitrary viewpoints and a simple map.

This paper proposes a linear algorithm for metric reconstruction from projective reconstruction. Metric reconstruction problem is equivalent to estimating the projective transformation matrix that converts projective reconstruction to Euclidean reconstruction. We build a quadratic form from dual absolute conic projection equation with respect to the elements of the transformation matrix. The matrix of quadratic form of rank 2 is then eigen-decomposed to produce a linear estimate. The algorithm is applied to three different sets of real data and the results show a feasibility of the algorithm. Additionally, our comparison of results of the linear algorithm to results of bundle adjustment, applied to sets of synthetic image data having Gaussian image noise, shows reasonable error ranges.

This paper proposes a new method to recover the sign of local Gaussian curvature from multiple (more than three) shading images. The information required to recover the sign of Gaussian curvature is obtained by applying Principal Components Analysis (PCA) to the normalized irradiance measurements. The sign of the Gaussian curvature is recovered based on the relative orientation of measurements obtained on a local five point test pattern to those in the 2-D subspace called the eigen plane. Using multiple shading images gives a more accurate and robust result and minimizes the effect of shadows by allowing a larger area of the visible surface to be analyzed compared to methods using only three shading images. Furthermore, it allows the method to be applied to specular surfaces. Since PCA removes linear correlation among images, the method can produce results of high quality even when the light source directions are not widely dispersed.

Recently a high-resolution image that has more than one million pixels is available easily. However, such an image requires much processing time and memory for an image understanding system. In this paper, we propose an integrated image understanding system of multi-resolution analysis and multi-agent-based architecture for high-resolution images. The system we propose in this paper has capability to treat with a high-resolution image effectively without much extra cost. We implemented an experimental system for images of indoor scenes.

We propose a new field of application for machine vision, a machine-vision-based cash-register system. We show that the overall system of color analysis for such an application should include the method of color distribution analysis which we propose, and that the analysis of shape and size is important. We present our test results and identify a few technical issues which may have to be considered for its practical utilization.

We have developed an airport monitoring system that traces the movement of airplanes in the parking areas of airports. For this system, we have developed an image processing method, a two-stage normalized background subtraction method that can detect moving objects and determine the sizes of those objects under illumination changes, which are inevitable for outdoor monitoring systems. The two-stage method consists of local and global normalized subtraction. With this method, airplanes can be detected in a stable manner under illumination changes, which means that the brightness in each pixel is not constant due to changes in atmospheric phenomena, such as the shadows of clouds. And false detection problems due to the presence of boarding bridges are solved by utilizing differences in motion between an airplane and the boarding bridge, such as the direction of movement. We have evaluated this method using 140 hours of video images that contain scenes with a variety of conditions, such as the presence of cloud shadows, the turning on and off of lights, night, rainfall and so on. As a result, we have confirmed a 95% level of accuracy of airplane detection. This system is now in operation at Kansai International Airport and is performing most satisfactorily.

In this paper, we describe a method for estimating external camera parameters in real time. We investigated the effectiveness of this method for annotating real scenes with 3-D virtual objects on a wearable computer. The proposed method enables determining known natural feature points of objects through multiplied color histogram matching and template matching. This external-camera-parameter calculation method consists of three algorithms for PnP problems, and it uses each algorithm selectively. We implemented an experimental system based on our method on a wearable vision system. This experimental system can annotate real objects with 3D virtual objects by using the proposed method. The system was implemented in order to enable effective annotation in a mixed-reality environment on a wearable computing system. The system consists of an ultra small CCD camera set at the user's eye, an ultra small display, and a computer. This computer uses the proposed method to determine the camera parameters. It then renders virtual objects based on the camera parameters and synthesizes images on a display. The system works at 10 frames per second.

This paper proposes a methodology for detecting matte-surfaced objects on a scene using color information and spatial thresholding. First, a difference image is obtained via a pixel-wise comparison of the color content of a 'clean' reference image and a sample image. Then, spatial thresholding of the difference image is performed to extract any objects of interest, followed by morphological post-processing to remove pixel noise. We study the applicability of two alternate color spaces (HSV, CIE Lab) for computing the difference image. Similarly, we employ two spatial thresholding methods, which determine the global threshold from the local spatial properties of the difference image. We demonstrate the performance of the proposed approach in scene surveillance, where the objective is to monitor a shipping dock for the appearance of needless objects such as cardboard boxes. In order to analyze the robustness of the approach, the experiment includes three different types of scenes categorized as 'easy,' 'moderate,' and 'difficult,' based on properties such as heterogeneity of the background, existence of shadows and illumination changes, and reflectivity and chroma properties of the objects. The experimental results show that relatively good recognition accuracy is achieved on 'easy' and 'moderate' scenes, whereas 'difficult' scenes remain a challenge for future work.

In this paper, we first introduce the notion of texture precision given the 3d geometry of a scene. We then provide an algorithm to acquire a texture/color map of the scene within a given precision. The texture map is obtained using projective devices (like pinhole sensing device) from data acquired either in the real world or computer-synthesized. Finally, we describe a procedure to obtain level of precisions by combining a modified edge-collapse geometry technique with an appropriate remapping texture engine. We report on our experiments and give perspectives for further research.

Recently, it becomes popular to synthesize new viewpoint images based on some sampled viewpoint images of real scene using technique of computer vision. 3D shape reconstruction in Euclidean space is not necessarily required, but information of dense matching points is basically enough to synthesize new viewpoint images. In this paper, we propose a new method for 3D reconstruction from three cameras based on projective geometry. In the proposed method, three input camera images are rectified based on projective geometry, so that the vertical and horizontal directions can be completely aligned with the epipolar planes between the cameras. This rectification provides Projective Voxel Space (PVS), in which the three axes are aligned with the directions of camera projection. Such alignment simplifies the procedure for projection between the 3D space and the image planes in PVS. Taking this advantage of PVS, silhouettes of the objects are projected into PVS, so that the searching area of matching points can be reduced. The consistency of color value between the images is also evaluated for final determination of the matching point. The finally acquired matching points in the proposed method are described as the surface of the objects in PVS. The acquired surface of the objects in PVS also includes knowledge about occlusion. Finally, images from new viewpoints can be synthesized from the matching points and occlusions. Although the proposed method requires only weak calibration, plausible occlusions are also synthesized in the images. In the experiments, images of virtual viewpoints, which were set among three cameras, are synthesized from three real images.

Functional magnetic resonance imaging (fMRI) allows to display functional activities of certain brain areas. In combination with a three dimensional anatomical dataset, acquired with a standard magnetic resonance (MR) scanner, it can be used to identify eloquent brain areas, resulting in so-called functional neuronavigation, supporting the neurosurgeon while planning and performing the operation. But during the operation brain shift leads to an increasing inaccuracy of the navigation system. Intraoperative MR imaging is used to update the neuronavigation system with a new anatomical dataset. To preserve the advantages of functional neuronavigation, it is necessary to save the functional information. Since fMRI cannot be repeated intraoperatively with the unconscious patient easily we tried to solve this problem by means of image processing and pattern recognition algorithms. In this paper we present an automatic approach for transfering preoperative markers into an intraoperative 3-D dataset. In the first step the brains are segmented in both image sets which are then registered and aligned. Next, corresponding points are determined. These points are then used to determine the position of the markers by estimating the local influence of brain shift.

We present a method for estimating activities of multiple, interacting objects detected by a video surveillance system. The activities are described in a stochastic context because our method is concerned with humans and uses noisy features detected from video. To monitor activities in this context, we introduce the concept of an attribute set for each blob, consisting of object type, action, and interaction. Using probabilistic relations introduced by a specific Markov model of these attribute sets, the activity descriptions are estimated from surveillance video.

We suggest the method to recover the 3D shape of an object by using a color image scanner which has three light sources. The photometric stereo is traditional to recover the surface normals of objects using multiple light sources. In this method, it usually assumes distant light sources to make the optical models simple. But the light sources in the image scanner are so close to an object that the illuminant intensity varies with the distance from the light source, therefore these light sources should be modeled as the linear light sources. In this method, by using these models and two step algorithm; the initial estimation by the iterating computation and the optimization by the non-linear least square method, not only the surface normal but also the absolute distance from the light source to the surface can be estimated. By using this method, we can recover the 3D shape more precisely. In the experimental results, the 3D shape of real objects can be recovered and the effectiveness of the proposed method is shown.

We describe a method for object recognition with 2D image queries to be identified from among a set of 3D models. The pose is known from a previous step. The main target application is face recognition. The 3D models consist of both shape and color texture information, and the 2D queries are color camera images. The kernel of the method consists of a lookup table that associates 3D surface normals with expected image brightness, modulo albedo, for a given query. This lookup table is fast to compute, and is used to render images from the models for a sum of square difference error measure. Using a data set of 42 face models and 1764 (high quality) query images under 7 poses and 6 lighting conditions, we achieve average recognition accuracy of about 83%, with more than 90% in several pose/lighting conditions, using semi-automatically computed poses. The method is extremely fast compared to those that involve finding eigenvectors or solving constrained equation systems.

Techniques for human-motion recovery are applicable to a variety of areas, such as sports, dancing, virtual reality, and video-game production. The people who work in this area focus their attention on recovering information on the motion of individuals rather than groups of people. It is important to demonstrate the possibility of recovering descriptions of the 3-D motion in team sports, since such information is able to provide us with a variety of information on the relations among players. This paper presents a new experimental result on 3-D motion recovery from a team sport. The result was obtained by a non-rigid shape recovery technique based on images from uncalibrated cameras. The technique was applied to recovering the 3-D motion of the players in a mini-basketball game which was played in a gymnasium. Some attention is focused on the analysis of the players' motion. Satisfactory results were obtained.

We have developed a novel human facial tracking system that operates in real time at a video frame rate without needing any special hardware. The approach is based on the use of Lie algebra, and uses three-dimensional feature points on the targeted human face. It is assumed that the roughly estimated facial model (relative coordinates of the three-dimensional feature points) is known. First, the initial feature positions of the face are determined using a model fitting technique. Then, the tracking is operated by the following sequence: (1) capture the new video frame and render feature points to the image plane; (2) search for new positions of the feature points on the image plane; (3) get the Euclidean matrix from the moving vector and the three-dimensional information for the points; and (4) rotate and translate the feature points by using the Euclidean matrix, and render the new points on the image plane. The key algorithm of this tracker is to estimate the Euclidean matrix by using a least square technique based on Lie algebra. The resulting tracker performed very well on the task of tracking a human face.

The problem of detecting straight lines arises frequently in image processing and computer vision. Here we consider the problem of extracting lines from range images and more generally from sets of three-dimensional (3D) points. The problem is stated as follows. Given a set Γ of points in 3D space and a non-negative constant , determine the line that is at a distance at most ε from the maximal number of points of . The extraction of multiple lines is achieved iteratively by performing this best line detection and removing at each iteration the points that are close to the line found. We consider two approaches to solve the problem. The first is a simple approach that selects the best line among a randomly chosen subset of lines each defined by a pair of edge points. The second approach, based on tabu search, explores a larger set of candidate lines thus obtaining a better fit of the lines to the points. We present experimental results on different types of three-dimensional data (i) range images of polyhedral objects (ii) secondary structures (helices and strands) of large molecules.

In order to maintain the diversity of structures in the population and prevent premature convergence, I have developed a new genetic algorithm called DCGA. In the experiments on many standard benchmark problems, DCGA showed good performances, whereas with harder problems, in some cases, the phenomena were observed that the search was stagnated at a local optimum despite that the diversity of the population is maintained. In this paper, I propose methods for escaping such phenomena and improving the performance by reinitializing the population, that is, a method called each-structure-based reinitializing method with a deterministic structure diverging procedure as a method for producing new structures and an adaptive improvement probability bound as a search termination criterion. The results of experiments demonstrate that DCGA becomes robust in harder problems by employing these proposed methods.

Deterministic execution testing has been considered a promising way for concurrent program testing because of its ability to replay a program's execution. Since, however, deterministic execution requires that a synchronization event sequence to be replayed be feasible and valid, it is not directly applicable to a situation in which synchronization sequences, being valid but infeasible, are taken into account. Resolving this problem is very important because a program may still meet its specification although the feasibility of all valid sequences is not satisfied. In this paper, we present a new approach to deterministic execution for testing concurrent systems. The proposed approach makes use of the notion of event independence and constructs an automation which accepts all the sequences semantically equivalent to a given event sequence to be replayed. Consequently, we can allow a program to be executed according to event sequences other than the given (possible infeasible) sequence if they can be accepted by the automation.

Disk-zoning technique has been widely adopted to increase disks capacities. As a result of disparity of capacities of inner and outer zones, the data transfer rates of the outer zones of a zoned-disk are higher than the inner zones that post a great challenge for zoned-disk based multimedia playback. In this paper, we study the data placement problem of VBR (variable bit rate) videos on zoned-disks. Our objective is to minimize video server buffer size and simultaneously to maximize disk utilization subject to the zone constraints of disk. We introduce the CRT (constant read time) method that allocates each user a constant time period in every service round to retrieve a variable-sized disk block. The CRT method can be formulated as constrained combinatorial problems that its optimum solution can be obtained by employing dynamic programming. Two heuristics are also explored to reduce time and space complexities. According to experiment results, the heuristic algorithms obtain near optimum solutions with shorter computation time.

This paper describes the adaptability of communication software through a biologically-inspired policy coordination. Many research efforts have developed adaptable systems that allow various users or applications to meet their specific requirements by configuring different design and optimization policies. Navigating through many policies manually, however, is tedious and error-prone. Developers face the significant manual and ad-hoc work of engineering an system. In contrast, we propose to provide autonomous adaptability in communication endsystem with OpenWebServer/iNexus, which is both a web server and an object-oriented framework to tailer various web services and applications. The OpenWebServer's modular architecture allows to abstract and maintain a wide range of aspects in a HTTP server, and reconfigure the system by adding, deleting, changing, or replacing their policies. iNexus is a tool for automated policy-based management of OpenWebServer. Its design is inspired by the natural immune system, particularly immune network, a truly autonomous decentralized system. iNexus inspects the current system condition of OpenWebServer periodically, measures the delivered quality of service, and selects suitable set of policies to reconfigure the system dynamically by relaxing constraints between them. The policy coordination process is performed through decentralized interactions among policies without a single point of control, as the natural immune system does. This paper discusses communication software can evolve continuously in the piecemeal way with biological concepts and mechanisms, adapting itself to ever-changing environment.

This paper treats the data routing problem for fault-tolerant systolic arrays based on Triple Modular Redundancy (TMR) in mixed spatial-temporal domain. The number of logical links required in TMR systolic array is basically 9 times larger than the one for corresponding non-fault-tolerant systolic array. The link sharing is a promising method for reducing the number of physical links, which may, however, degrade the fault tolerance of TMR system. This paper proposes several robust data-routing and resource-sharing (plural data transfers share a physical link, or a data transfer and a computational task share a PE as a relay node for the former and as a processor for the latter), by which certain classes of fault tolerant property will be guaranteed. A stage and a dominated set are introduced to characterize the features of routing/resource-sharing in TMR systems, and conditions on the dominated set and their resultant fault-tolerant properties are derived.

A mesh-connected processor array consists of many similar processing elements (PEs), which can be executed in both parallel and pipeline processing. For the implementation of an array of large numbers of processors, it is necessary to consider some fault tolerant issues to enhance the (fabrication-time) yield and the (run-time) reliability. In this paper, we introduce the 1(1/2)-track switch torus array by changing the connections in 1(1/2)-track switch mesh array, and we apply our approximate reconfiguration algorithm to the torus array. We describe the reconfiguration strategy for the 1(1/2)-track switch torus array and its realization using WSI, especially 3-dimensional realization. A hardware realization of the algorithm is proposed and simulation results about the array reliability are shown. These imply that a self-reconfigurable system with no host computer can be realized using our method, hence our method is effective in enhancing the run-time reliability as well as the fabrication-time yield of processor arrays.

In content-based image retrieval (CBIR), the content of an image can be expressed in terms of different features such as color, texture, shape, or text annotations. Retrieval methods based on these features can be varied depending on how the feature values are combined. Many of the existing approaches assume linear relationships between different features, and also require users to assign weights to features for themselves. Other nonlinear approaches have mostly concentrated on indexing technique. While the linearly combining approach establishes the basis of CBIR, the usefulness of such systems is limited due to the lack of the capability to represent high-level concepts using low-level features and human perception subjectivity. In this paper, we introduce a Neural Network-based Image Retrieval (NNIR) system, a human-computer interaction approach to CBIR using the Radial Basis Function (RBF) network. The proposed approach allows the user to select an initial query image and incrementally search target images via relevance feedback. The experimental results show that the proposed approach has the superior retrieval performance over the existing linearly combining approach, the rank-based method, and the BackPropagation-based method.

An improved algorithm is developed for correcting the topographic impact on satellite imagery. First, we analyze the topography induced distortion on satellite image. It is shown that the variation of aspect can cause the obvious different distortions in the remotely sensed image, and also effect the image illumination significantly. Because the illumination is the basis for topographic correction algorithms, we consider its variation in different sun-facing aspects in calculation a correction parameter and take it as a key element in the modified correction algorithm. Then, we apply the modified correction method on the actual Landsat Thematic Mapper satellite image. The topographic correction was done in different image data with different season and different solar angle. The corrected results show the effectiveness and accuracy using this approach.

Wavelet based image compression is getting popular due to its promising compaction properties at low bitrate. Zerotree wavelet image coding scheme efficiently exploits multi-level redundancy present in transformed data to minimize coding bits. In this paper, a new technique is proposed to achieve high compression by adding new zerotree and significant symbols to original EZW coder. Contrary to four symbols present in basic EZW scheme, the modified algorithm uses eight symbols to generate fewer bits for a given data. Subordinate pass of EZW is eliminated and replaced with fixed residual value transmission for easy implementation. This modification simplifies the coding technique as well and speeds up the process, retaining the property of embeddedness.

Footprint assembly was proposed to reduce the blurriness of texture mapped image by mipmapping. Even though it can improve the quality of texture mapped image, there are yet blurring due to the limitation of it's filter kernel. This paper proposes a novel texture filtering, called adaptive footprint assembly (AFA), to overcome the limitation of footprint assembly. The proposed method greatly improves the quality of texture mapped images.