A new approach of data clustering which is capable of detecting linked or crossed clusters, is proposed. In conventional clustering approaches, it is a hard work to separate linked or crossed clusters if the cluster prototypes are difficult to be represented by a mathematical formula. In this paper, we extract the force information from data points using the concept of psychological potential field, and utilize the information to measure the similarity between data points. Through several experiments, the force shows its effectiveness in diiscriminating different clusters even if they are linked or corssed.

The warp model of the extended Hough transform (EHT) has been proposed to design the explicit expression of the transform function of EHT. The warp model is a skewed parameter space (R(µ,ξ), φ(µ,ξ)) of the space (µ,ξ), which is homeomorphic to the original (ρ,θ) parameter space. We note that the introduction of the skewness of the parameter space defines the angular and positional sensitivity characteristics required in the detection of lines from the pattern space. With the intent of contributing some solutions to basic computer vision problems, we present theoretically a dynamic and centralfine/peripheral-coarse camera vision architecture by means of this warp model of Hough transform. We call this camera vision architecture askant vision' from an analogy to the human askant glance. In this paper, an outline of the EHT is briefly shown by giving three functional conditions to ensure the homeomorphic relation between (µ,ξ) and (ρ,θ) parameter spaces. After an interpretation of the warp model is presented, a procedure to provide the transform function and a central-coarse/peripheralfine Hough transform function are introduced. Then in order to realize a dynamic control mechanism, it is proposed that shifting of the origin of the pattern space leads to sinusoidal modification of the Hough parameter space.

This paper describes a new feature extraction model (Active Model) which is extended from the active contour model (Snakes). Active Model can be applied to various energy minimizing models since it integrates most of the energy terms ever proposed into one model and also provides the new terms for multiple images such as motion and stereo images. The computational order of energy minimization process is estimated in comparison with a dynamic programming method and a greedy algorithm, and it is shown that the energy minimization process in Active Model is faster than the others. Some experimental results are also shown.

The Paper describes a comprehensive system for image recognition based on the technique of boundary spline matching. It can be used to accurately compare two objects and determine whether they are identical or not. The result is extremely satisfactory for comparing planar objects as revealed from the illustrative example presented in this paper. In real practice, images of the same scene object can easily be considered as belonging to different objects if the objects are viewed from different orientations and ranges. Thus, image recognition calls for choosing the proper geometric transformation functions to match images as the initial step so that recognition by template matching can be done as the second step. However, there are a large variety of transformation functions available and the subsequent evaluation of transformation parameters is a highly nonlinear optimisation procedure which is both time consuming and not solution guaranteed, making real-time estimation impossible. This paper describes a new method that represents the boundary of each of two image objects by B-splines and matches the B-splines of two image objects to determine whether they belong to the same scene object. The algorithm developed in this paper concentrates on solving linear simultaneous equations only when handling the geometric transformation functions, which takes almost negligible computational time by using the standard Gaussian Elimination. Representation of the image boundary by B-splines provides a flexible and continuous matching environment so that the level of accuracy can be freely adjusted subject to the requirement of the user. The non-linear optimisation involves only one parameter, i.e. the starting point of each boundary under B-spline simulation, thus guaranteeing a very high speed computational system. The real time operation is deemed possible even there is a wide choice of proper transformation functions.

An active contour model which is called Snakes was proposed to extract the border line of an object from an image. This method presents the minimization problem of the energy function defined on the contour curve. The authors obtained an excellent result by applying genetic algorithm to the contour extraction. In this paper, the biphased genetic algorithm, which is a new type of genetic algorithm, is proposed to minimize the energy function of Snakes. The parameters of the genetic algorithm are examined to tune up its local and global search abilities. The biphased genetic algorithm composed of two phases of genetic search is constructed to use both abilities of the exploration and the exploitation properties of the genetic algorithm. The processing results of the biphased genetic algorithm are compared with those of the previous methods, and the advantages of the proposed algorithm are shown by several experiments.

Introducing a general statistical model of image noise, we present an optimal algorithm for computing 3-D motion from two views without involving numerical search: () the essential matrix is computed by a scheme called renormalization; () the decomposability condition is optimally imposed on it so that it exactly decomposes into motion parameters; () image feature points are optimally corrected so that they define their 3-D depths. Our scheme not only produces a statistically optimal solution but also evaluates the reliability of the computed motion parameters and reconstructed points in quantitative terms.

We prove that the maximum likelihood estimator for estimating 3-D motion from noisy optical flow is not optimal", i.e., there is an unbiased estimator whose covariance matrix is smaller than that of the maximum likelihood estimator when a Gaussian noise distribution is assumed for a sufficiently large number of observed points. Since Gaussian assumption for the noise is given, the maximum likelihood estimator minimizes the mean square error of the observed optical flow. Though the maximum likehood estimator's covariance matrix usually reaches the Cramér-Rao lower bound in many statistical problems when the number of observed points is infinitely large, we show that the maximum likelihood estimator's covariance matrix does not reach the Cramér-Rao lower bound for the estimation of 3-D motion from noisy optical flow under such conditions. We formulate a superior estimator, whose covariance matrix is smaller than that of the maximum likelihood estimator, when the variance of the Gaussian noise is not very small.

This paper presents a modified disparity measurement to recover the depth and a robust method to estimate motion parameters. First, this paper considers phase correspondence for the computation of disparity. It has less computation for disparity than previous methods that use the disparity from correspondence and from correlation. This modified disparity measurement uses the Gabor filter to analyze the local phase property and the exponential filter to analyze the global phase property. These two phases are added to make quasi-linear phases of the stereo image channels which are used for the stereo disparity finding and the structure recovery of scene. Then, we use feature-based correspondence to find the corresponding feature points in temporal image pair. Finally, we combine the depth map and use disparity motion stereo to estimate 3-D motion parameters.

In this paper, we present an approach which is applicable to both the stereo and the motion correspondence problems. We take into account different representations of edge primitives and introduce the idea of Hough Transform to develop a matching algorithm which does not require any local constraints during the matching process.

For medical imaging, non-rigid motion analysis of the heart deformability is a nontrivial problem. This paper introduces a new method to analyze the gated SPECT (Single Photon Emission Computed Tomography) imges for 3-D motion information of left ventricular. Our motion estimation method is based on a new concept called normal direction constraint" in that the normal of a surface patch of some deforming objects at certain time instant is constant. This paper consists of the following processes: contour extraction, slices interpolation, normal vector field generation, expanding process, motion estimation for producing a 2-D motion vector field, and deprojection for a 3-D motion vector field. In the experiments, we will demonstrate the accuracy of our method in analyzing the 3-D motion field of deforming object.

This paper describes a new method for pose estimation of human face moving abruptly in real world. The virtue of this method is to use a very simple calculation, disparity, among multiple model images, and not to use any facial features such as facial organs. In fact, since the disparity between input image and a model image increases monotonously in accordance with the change of facial pose, view direction, we can estimate pose of face in input image by calculating disparity among various model images of face. To overcome a weakness coming from the change of facial patterns due to facial individuality or expression, the first model image of face is detected by employing a qualitative feature model of frontal face. It contains statistical information about brightness, which are observed from a lot of facial images, and is used in model-based approach. These features are examined in everywhere of input image to calculate faceness" of the region, and a region which indicates the highest faceness" is taken as the initial model image of face. To obtain new model images for another pose of the face, some temporary model images are synthesized through texture mapping technique using a previous model image and a 3-D graphic model of face. When the pose is changed, the most appropriate region for a new model image is searched by calculating disparity using temporary model images. In this serial processes, the obtained model images are used not only as templates for tracking face in following image sequence, but also texture images for synthesizing new temporary model images. The acquired model images are accumulated in memory space and its permissible extent for rotation or scale change is evaluated. In the later of the paper, we show some experimental results about the robustness of the qualitative facial model used to detect frontal face and the pose estimation algorithm tested on a long sequence of real images including moving human face.

Based on a newly proposed notion of relational network, a novel learning mechanism for model acquisition is developed. This new mechanism explicitly deals with both qualitative and quantitative relations between parts of an object. Qualitative relations are mirrored in the topology of the network. Quantitative relations appear in the form of generalized predicates, that is, predicates that are graded in their validity over a certain range. Starting from a decomposition of binary objects into meaningful parts, first a description of the decomposition in terms of relational networks is obtained. Based on the description of two or more instances of the same concept, generalizations are obtained by first finding matchings between instances. Generalizing itself proceeds on two levels: the topological and the predicate level. Topological generalization is achieved by a simple rule-based graph generalizer. Generalization of the predicates uses some ideas from MYCIN. After successful generalization, the system attempts to derive a simple and coarse description of the achieved result in terms of near natural language. Several examples underline the validity of relational networks and illustrate the performance of the proposed system.

In general, message passing multiprocessors suffer from communication overhead and shared memory multiprocessors suffer from memory contention. Also, data I/O overhead limits performance. In particular, computer vision tasks that require massive computation are strongly affected by these disadvantages. This paper proposes new parallel architectures for computer vision, a Flexibly (Tightly/Loosely) Coupled Multiprocessor (FCM) and a Flexibly Coupled Hypercube Multiprocessor (FCHM) to alleviate these problems. FCM and FCHM have a variable address space memory in which a set of neighboring memory modules can be merged into a shared memory by a dynamically partitionable topology. FCM and FCHM are based on two different topologies: reconfigurable bus and hypercube. The proposed architectures are quantitatively analyzed using computational models and parallel vision algorithms are simulated on FCM and FCHM using the Intel's Personal SuperComputer (iPSC), a hypercube multiprocessor, showing significant performance improvements over that of iPSC.

In this paper, we present a digital scheme for fast VLSI range imaging sensor, which is a modification of the analog scheme of existing sensor implemented by T. Kanade. Instead of reading timing information in analog manner, we use a digital scheme which has several advantages over the analog scheme, including area saving, insusceptibility to noise and other undesirable effects. We have implemented a prototype to test feasibility and present its experimental result.

In this letter we propose an interpolation technique for low-quality fingerprint images for highly reliable feature extraction. To improve the feature extraction rate, we extract fingerprint features by referring to both the interpolated image obtained by using a directional Laplacian filter and the high-contrast image obtained by using histogram equalization. Experimental results show the applicability of our method.