We have developed an experimental 4K x 2K pixel progressive scan color camera system with a four-sensor pickup method in which two sensors are used for green and the rest are for red and blue. The diagonal spatial offset imaging method has been applied to the two green sensors. By interpolating their output signals, the number of vertical and horizontal pixels of the green channel increases to twice that of one imager. For red and blue, a line-offset imaging method is applied between red and blue signals. This means that the sample point of red and blue signals is half that of the green signals. In this paper, we report on our investigations into the alias signal suppression on red and blue signals by using optical low-pass-filter and signal interpolations.

View full abstract

Coarseness and directionality provides important sources of information for color texture image recognition. Especially, it is important to distinguish between the textures and understand characters of similar color texture. So we proposed new scale-space analysis generated by non-uniform Gaussian kernels in order to find stabile image for coarseness and directionality. We analyze zero-crossing surfaces to generate non-uniform Gaussian scale-space from observations of a limited number. Singular points, where the topology of zero-crossing surfaces changes are plotted in new scale-space. A filter parameter for the biggest size of chunk enclosed by topology change surface is selected as an optimal parameter of a pixel. Optimal filter and the image description are calculated by this approach for natural color image. We show that this method is suited for color texture image recognition.

View full abstract

This paper describes "CyberCreater" that enables general authors to construct space for providing services easily in a virtual space system for multiple users calle "InterSpace" "CyberCreater" enables general authors to construct space for providing services and define services to provide with user interface which supports author's intuitional understanding. Further, it supports communication environment among users, information and services by making visual agents possible.

View full abstract

In penumbral imaging, the noise contained in encoded image will significantly affect decodede image. In this paper we proposed to use an adaptive windowing method to filter the noise of the encoded images, followed by conventional decoding method. The proposed adaptive windowing method can filter the noise with any blurriog of the edge. The applicanbility of the proposed method for penumbral imaging has been demonstrated by computer simulations.

View full abstract

Usually a single-pinhole camera is used for X-ray imaging. The limitation of the single-pinhole camera is that it can not combine good collection efficiency with high resolution. In URA camera, the pinhole is replaced by multi-pinholes arrays arranged in m-sequences. So the URA camera can provide an image with a high resolution and high SN ratio, and image distant planar sources with no artifacts. Additional advantage. URA camera has tomographical properties for 3-D depth objects. In this paper we investigate performance of the URA camera for tomographic imaging.

View full abstract

We propose an algorithm that can reduce very huge processing of two-dimensional correlation which is used for the block matching between spatially differentiated images. The algorithm consists of a sequential processing of Hough transform, the one-dimensional convolution and correlation along the ρ coordinate of Hough plane, and an inverse Hough transform.

View full abstract

Multiplication-like operations for correlation function have been proposed. The operations produce the output with the same polarity as the product of the two inputs when the product is not zero and give zero when it is zero. The operations needs smaller processing cost than multiplication if it is adequately arranged. We discuss the implementations of the operation la+bl-la-bl that is one of the most promising operation among the multiplication-like operations. We also discuss a general method that gives multiplication-like operations.

View full abstract

Much research has been done on how to change gray-level image to binary image. We propose a method of binarization for simple figures on character images whichare input with CCD camera. The optimum threshold which depends on images is decided by the dynamic thresholding, and the equation is derived by the correlation between experimentally obtained optimum threshold and gray-level of background segment on an image. This method has an advantage that we can get binary image from an original image even with noise of shading more easily than conventional ways.

View full abstract

The image restoration is to recover the original image from the degraded image. If the blurring function is known, the original image can be recovered by using some linear deconvolution techniques such as inverse filter and Wiener filter. But in many cases of image processing, it is difficult to know the exact blurring function. The recovery of original image from the knowledge of degraded image alone is known as the blind deconvolution problem. It is impossible to restore image by using the conventional linear deconvolution techniques because the blurring function is unknown. In this paper, we proposed to use a genetic algorithm (GA) to solve the blind deconvolution problem.

View full abstract

A new simulator for MPEC-1 encoder with interactive manual operating tools has been developed. The operator of this simulator can select the type of preprocessing filters and the quality level for each segmented region. A large amount of code is allocated to significant regions (ex. person, important subject). However in unsignificant regions (ex. background), high frequency components are removed using low pass filter and a small amount of codes is allocated. Effectiveness of this simulator is verified by computer simulation.

View full abstract

In this paper, we discuss the communication environment in multi-user shared three dimensional cyberspace. The major problems in the environment are two points as follows; 1) Appropriate selection of information for promoting communication under the limited display performance. 2) Coexistence of two methods for communication. One is the method for obtaining crowd information, the other is the method for person-to-person communication in such a situation. This paper proposes the efficient display method of communication environment along with the simulation results.

View full abstract

Multimedia becomes popular in receut years and dealing with a motion picture on computersis getting usual. And now MPEG is generaly used as a international standard in case of dealing with a motion picture on compnters. We detected motion regions by applying a motion vector encoded in MPEG data without decoding MPEG data. After that we detected contorlines of motion regions. The resolution is limitted to macro blocks. But we suppose it enongh for practical use. And by applying it we propose a 3-dimensional motion vectors extracting method. We interpret the motion vectors acquired from two or more images of a same object taken from different viewpoints at the same time and get 3-dentensional motion vector.

View full abstract

The purpose of this experiment is to analyze the relation between pupil and heart rate under active and passive stress. The changes of pupillary diameter and heart rate variability were measured under three conditions : with no, mental and physical stress.The results showed that mental stress decreased pupillary hippus and physical stress increased the pupil size, which suggests that different types of stress cause different features of pupillary dynamics.

View full abstract

This study proposed first-passage-time function models of one-dimensional Brownian motion with drift for spontaneous eye blinking, and investigated blink burst at low level of vigilance with the models. The first-passage-time probability density was calculated with two parameters : distance from the initial point to the threshold and the drift rate, and analyzd with Pearson plots. The results showed that the decrease of distance to the threshold caused blink burst which exhibited exponential function distribution in joint histogram, but the shift of drift rate caused no changes in the distribution.

View full abstract

Wavelet analysis of eye movements is proposed. This method enables us to get low noise waveform of eye movement from raw data, therefore, you can get easily time derivatives of the data, i.e. angular veiocity and angular acceleration. These are basic quantities that you can get clear structure of eye movements as a dynamical system. This will be a basic evaluation method of picture quality of novel visual displays using virtual images and stereoscopic images. And also this method will be applied to visual fatigue problems for various kinds of visual displays.

View full abstract

The differences in motion-detection characteristics between the Fourier and the non-Fourier motion-detection mechanisms, were examined based on some experimental results. In these experiments subjects were instructed to answer the perceived motion direction to presented apparent motion stimuli with various kinds of inter-stimulus intervals (ISIs) during which the field was spatially uniform with Iuminance equal to the average luminance of the stimulus field. Experimental results gave us the following suggestions. (1) In two-frame apparent motion, ISIs make the Fourier mechanism produce perceptual motion in the direction opposite to the physical displacement (reverse motion). (2) In successive apparent motion, ISIs cause the Fourier mechanism not to operate. (3) The non-Fourier mechanism produces perceptual motion in the direction of the physical displacement regardless of ISIs.

View full abstract

The factor analysis by means of the technicque of semantic differential was applied to making the relationship between different modes of color displays clear. It was found that three factors, "beauty", "activity" and "weight" were commonly appeared but in different foctor loadings and that others, "reality", "texture" and "sharpness" were also effective to characterize each display.

View full abstract

On shooting 3D-HDTV, the inconsistency of the spatial relationship between L/R cameras or of the focal length of lenses results in geometrical distortions between L/R images such as vertical shift, size inconsistency and rotation error. We studied the acceptable margin of these simultaneous distortions by multiple regression analysis. Subjective evaluation tests showed that the estimation of the margin is more precise with the shift of the corresponding points on L/R images caused by these distortions, rather than that with these distortions themselves. The partial regression coefficients show that the vertical shift has more influence than the horizontal shift.

View full abstract

We investigated the relationship between the perception of absolute distance and the absolute disparity, which the pictures presented in a stereoscopic display had. We presented the stereo pictures with uniform disparity to the subjects and measured their perceived distances to the display for investigating the relationship quantitatively. The results in experiment 1 showed that the perception of absolute distance depended on the absolute disparity for two subjects out of eight, but the perceived distance was independent of the absolute disparity for other six subjects. The results in experiment 2 showed that when the test part of the display having uniform disparity was presented with a reference part having zero disparity, the absolute distance for the test depended on the test disparity for the subjects who showed little effect for the display with no reference in experiment 1,but the effect of the disparity on the absolute distance decreased for the subjects who showed some effects in experiment 1.

View full abstract

This paper presents a real-time visual tracking system that can follow a moving object by controlling the pan and tilt of a camera based on the object position and speed obtained by video-rate image processing. The system consists of a camera, an image processor, a pan/tilt platform and a motor driver. The developed platform can control the pan and tilt angle of a camera with accuracy of 0.002 [deg] and maximum speed of 50 [deg/s]. We have developed a tracking algorithm using both position and velocity of a moving object. The system integrates them to capture a moving object in the center of camera view. The algorithm, implemented on a general purpose image processor, can detect a moving object within 33 [ms]. As an experimental result, the system can track a moving object (walking person) in real-time.

View full abstract

We are studying about an intelligent robot camera that can shoot convincing real images automatically, for broadcasting use. For this purpose, camera work of broadcasting cameramen have been analyzed. In this paper, the condition which cameramen were shooting the moving players in soccer games was focused. Here are the findings ; occurrence ratio of gazing velocity 50 msec is higher, cameramen shoot the subjects keeping space in the image toward the subjects' moving direction ; in the case of the players' quick movement, cameramen controls the camera movement within about 300 msec ; etc.

View full abstract

The task of word sense disambiguation using electronic dictionary is to choose one of the definitions associated the entry word. An associative network constructed by entry word's node and word's node of definition in the electronic dictionary, we conducted an experiment to disambiguate word senses of test data selected from the standard norm of word association. As a result, it has shown that over 70 % of correct disambiguation can be achieved against about 20 % correct disambiguation by random choice.

View full abstract

In this paper, we present feedforward neural network with an automatic gain amplifier in the output layer for solving the scaling problem. We also propose an adaptive learning algorithm for updating the gain factor m(t), so the any range of desired output can be achieved without a prior knowledge of unknown nonlinear system. Due to adaptive gain factor, the high estimation accuracy and fast convergence can be obtained. Simulation results are presented to illustrate the effecteveness and performance of the proposed method.

View full abstract

In recurrent neural network (RNN) learning of finite state automata (FSA), we discuss how a neruo gain (β) influences the stability of the state representation and the performance of the learning. We formally show that the existence of the critical neuro gain (β0) : any β larger than β0 makes an RNN maintain the stable representation of states of an acquired FSA. Considering the existence of β0 and avoidance of local minima, we propose a new RNN leaning method with the scheduling of β, called an annealed RNN learning. Our experiments show that the annealed RNN learning went beyond a constant β learning.

View full abstract

We consider the problem of inferring invariant measures of dynamical systems. To solve the problem, we give a new class of ergordic mappings whose invariant measure is absolutely continuous with respect to the Lebesgue measure using the addition formula of hyper-elliptic functions. In this class of ergordic dynamical systems, this inverse problem is shown to be transformed to be the statistical inference problem in the standard sense.

View full abstract

Modified Hebbian learning rule (Griniasty et al. 1993) or trace mechanism (Foldiak 1992) is one account for the problem of how the temporal relations among events are represented in neural connections. Here we show another possibility to consider random encoding strategy in which items to be stored are associated with the randomly chosen attractors of a network independent of spatial relationships between them. Namely, the network spontaneously generate random patterns of activity as internal representations for events or objects. This was demonshated by the network models in which connection weights are i.i.d.random variables under the symmetric condition. Mathematical properties on this system, such as distances between randomly chosen attractors, methods generating the random patterns and the dependences of the methods, are theoretically considered. Lastly, we discuss its relevances to the physiological evidences for memory.

View full abstract

Proposed in this paper is a novel neural networks approach to inverse optimization problems with nonstandard quadratic criterion functions. An inverse optimization problem here means to obtain a positive semidefinite quadratic criterion function which makes a given solution optimal under given constraint conditions. However, in contrast to the case of standard quadratic criterion functions, a cruterion function cannot be determined uniquely even when there is only one active constraint condition. To solve this difficulty, it is proposed to obtain a criterion function closest to the studard quadratic criterion function. For this purpose, the sum of the absolute values of off-diagonal connection weights and that of squares of them are used as indicators for the distance from the standard quadratic criterion function. A structural learning with forgetting applied to off diagonal connection weights successfully generates a simplified quadratic criterion function. How criterion parameters and Lagarange multipliers changes for various gradient vectors is also analyzed.

View full abstract

Pattern recognition invariant against distortion and shift can be performed with the dynamic link architectures. The quality of a match between two images is assessed through an clastic cost function. According to the definition of the cost function, this model is not invariant against rotation and expansion. We extended the dynamic link architectures to a continuous model based on a local linear map which is expressed by partial differential equations. Our model produce a topographic mapping between two images by an interaction between two systems, the global map and the local linear map. Therefore our model is very flexible against distortion and rotation compared with the dynamic link architectures

View full abstract

All preceeding models of self-organization of topographic mappings have been based on selfsustaining local excitation in nerve fields. We proposed a new model using nerve fields which allow excitatory travelling waves and showed the formation of topographic mapping in the model. We can understand this model, as well as previous models, from the viewpoint of phase-locking between excitation patterns in two nerve fields interacting through the connection. It is different from any previous models in that the formation of the mapping depends on the spatio-temporal structure of the excitation patterns on the presynaptic nerve field. By decreasing the width of excitation gradually, computer simulations have shown the model to be able to form almost complete topographic mapping from initial values made by uniformly distributed random numbers.

View full abstract

Aonishi et al. previously proposed the eye movement model based on non-uniformity of the receptive fields in visual systems. Their model was designed for handling only line drawing on a white background. We propose a new model that can handle half-toned images taken by a CCD camera. We modified the ganglion cells that detect contrast from the input image, and feature extractioning unit which extract edges and bend-of-lines. By this modification, our model can scan these feature points on half-toned images.

View full abstract

The sensitivities of the geometrical characteristics of arm stiffness at hand during force control in the horizontal plane were investigated, which are indexed by shape and orientation of stiffness ellipse and characterized by arm posture and ratios between shoulder, elbow, and double-joint stiffness. At the distal hand position, as previously studied, geometrical characteristics, shape and orientation of stiffness ellipse, derived from arm stiffness matrix, were not much altered under different requested conditions. At the proximal hand position, those characteristics, however, were much altered. From theoretial analyses, it has been revealed that the shape and orientation of ellipse are insensitive to the changes of the joint stiffness ratios at the distal hand position, and are sensitive to those changes at the proximal hand position. In other words, slight changes of geometrical characteristics of stiffness ellipse do not directly indicate slight changes of joint stiffness ratios.

View full abstract

A Self-organizing Information Retrieval (IR) System, named Sir-Web, is being developed to contribute to a Self-organized Information Data Base Research Project. Sir-Web is designed to be useful in the future information oriented society especially to allow novice user to access information references easily. The developmental goals of Sir-Web are summarized as follows : 1. Sir-Web is automatically constructed without an IR database generated by IR experts. 2. Sir-Web has a user-friendly graphic interface that supports "Visualization of Infomation Retrieval Process" for novice users. 3. Sir-Web has the ability to learn through use and evaluation. and this ability makes Sir-Web adaptive and smart. 4. Sir-Web works as a distributed Information Retrieval system on the Internet.

View full abstract

We have developed a system for translation and speech from common Japanese to Ryukyu dialect that works on WWW. Pursuit of translation accuracy and high-quality speech is now running. The system is open to the public on WWW by using hyper-text description. The system translate standard Japanese text to Ryukyu dialect, then it speaks the dialect text by speech synthesis. In this paper, we describe an outline of the system and its operation.

View full abstract