This paper considers a segmentation algorithm of the character string from the color image. For this purpose, we use the 3-D tensor product expansion which represents the image as a product of a number of vectors. Simulations are carried out, and we confirm the validity of the method.

View full abstract

In this paper, first, a simple 2-D IIR filter is considered based on the Roesser local state-space model. Applying this filter to image processing from four directions, it is possible to carry out smoothing, edge detection and edge emphasis without any distortion. Next, for effective noise reduction, two hybrid filters are considered. These are the combination of both IIR filters and nonlinear filters, like median filters and nueral filters. Each method has several merits about an amount of computation, flexibility and sharpness. Finally, some examples are given to illustrate the utility of the proposed technique.

View full abstract

A restoration method for shift variant blurred image estimating the parameter distribution of PSF is described. In this study, it is assumed that the blurred image is degraded by either the defocus blur or the uniform linear motion blur. Accordingly, the estimation problem of the PSF is reduced to the estimation of the parameter distributions, which are the diameter of the circular disk of the defocus and the width of the motion. A number of restored images obtained by the conventional Wiener filters corresponding to a number of blur parameters of the PSF are used for the estimation of the parameter distribution and the adaptive restoration of the shift variant blurred image. The results of computer simulation are shown for demonstrating the efficacy of the proposed restoration algorithm.

View full abstract

This paper describes a texture segmentation algorithm using LOG-filters and stochastic relaxation. The goal is to find the distinct texture regions by assigning categorical labels to the pixels. First, multiresolution edges in a texture image is detected by LOG-filters. Texture features are local density of the detected edges. Secondly, assigning categorical labels are driven by the detected feature values of each pixel. The spatial constraints are enforced through the use of a stochastic relaxation algorithm that includes global interactions. Finally, experimental results show the effectiveness of the segmentation algorithm.

View full abstract

Computing scales for multiresolutional LOG filters is discussed. One of the problems in the Abstract filter type edge detection is how to select scales of the filters. In previous research, the scales were always given by the operators. In this paper, we propose a method for computing six scales based on the multi-channel theory in human vision and selecting a suitable scale for the target image by comparing a power spectrum of the image with a mean power spectrum of sample images. Natural edges are detected by the filters scaled with the proposed method.

View full abstract

The method of convex projections is a recursive algorithm to restore the original image in the intersection of a closed convex sets preassigned from every known property of an original image employing only nonlinear projection operators. Except for especial cases, the algorithm gives only a point of intersection in the sense of weak-convergence. Moreover, the weakly converged point depends on the permutation of projection operators. To cope with their short comings, we propose a new recursive algorithm using Browder strongly convergence theorem.

View full abstract

A coding control scheme by an adaptive pre-filter is proposed. Optimum point in quality is analyzed in constant bit rate codig considering quantization noise and loss of sharpness by two factors; coding output amount given by both quantization step size and bandwidth of input picture, and a balance between those elements. The proposed pre-filter changes its property adaptively for dynamically optimizing the picture quality refering to some coding characteristics. Some computer simulations show that it compares favorably with the coding control only by quantization step size in respective of the subjective picture quality especially for scenes with large motion.

View full abstract

This paper analyzes data distortion which comes from 2:1 resolution conversion with DCT Scaling decoder which expands a part of compressed data. The analysis leads an objective measure for system evaluation, including not only amplitude but also phase, to compare some down-scaling methods and to discuss phase shift problem. The measure is driven from treating the DCT scaling decoder as a filter bank and calculating difference between input and output signal of the system in frequency domain.

View full abstract

Accurate detection of pixel-level motion vectors is very important in video coding and video processing. In this paper, a new method of detecting pixel-level motion vectors is proposed. It is shown that pixel-level motion. vectors can be detected by a block matching method. It makes use of hierarchical images to detect motion vectors. Hierarchical images are obtained from an image contracted by subsampling alternately and from the image expanded by interpolation. It is shown that accuracy further improves by referring to adjacent blocks. Efficacy of the proposed method is confirmed by simulation.

View full abstract

This paper describes some techniques of predictive coding using look-up table. Quantizer input in an ordinary DPCM coder is a prediction error. the prediction error distribution depends on the predicted value, and the range of prediction error values is bounded by the predicted value. If different quantizers are switched depending on the predicted value, the reconstruction error could be made much smaller. The quantizer being switched by predicted value is useful in avoiding transmission error propagation. This effect is explained by an illustration of code allocation. Some other features of predictive coders are also illustrated.

View full abstract

In this paper, we discuss VBR transmission of video with constraints on the average rate considering the usage paramter control in an ATM network. To begin with, we propose a method to keep the average rate under a specified value. Then we compare the video transmission with the proposed rate control method with CBR transmission and VBR transmission using constant quantizer step size, in terms of video quality, cell loss probabilty and duration of congestion. We also examine how the interval to calculate average rate affects these respects. It is shown that, the video transmission with the proposed method shows considerably better video quality in scene changes than the CBR transmission and shows signifficantly better performance of statistical multiplexing than the VBR transmission using constant quantizer step size. It is also shown that extending the interval to calclulate the average rate makes video quaity more stable but, when the number of multipelxed video sources is small, this results in higher cell loss ratio and longer congestion.

View full abstract

A programmable video signal processing system for HDTV (HD-Picot system) has been developed, which carries out image processing functions, including spatio-temporal filtering, motion-vector detection, contour detection, color processing, pattern matching, geometrical transformation, image synthesis, texture mapping, on NTSC and HDTV signals at video-rate. HD-Picot system is a multi-processor composed of cascaded clusters which includes homogeneous 16 processors connected with network. The system realizes performance in propotion to the number of processors, because (1) it avoids memory-access conflict by giving image memory to each processor, (2) it achieves both pipeline and parallel processing by employing network to permit all connections between processors, and (3) it adopts sophisticated program-control system to synchronize with sync signals.

View full abstract

On double-rate scanning systems of Hi-Vision whose scanning parameter is 1125/60/2:1, characteristics of a camera, degree of interlace artifact and picture quality are investigated. The double-rate systems are "Pro-scan (1125/60/1:1)", "120/2:1 (1125/120/2:1)", As results, relation between scanning "2250(2250/60/2:1)", etc. parameters and spatio-temporal resolution of a Saticon camera become clear. In temporal resolution, there is the greatest difference of resolution among scanning systems. As regards interlace artifact, it become clear that the artifact is improved by doubling the field rate of systems. Results of subjective evaluation of impairments by line structure, resolution and picture quality are showed for these systems.

View full abstract

Digital video-on-demand service provided by a digital network using video compression techniques will require visual search functions such as fast forward and fast reverse, which increase the load on storage devices, to be performed with multiple access. This paper therefore proposes new techniques for multiple-access fast forward and fast reverse, for flexible visual search on a frame-by-frame basis, and for quick switch-over between different visual search functions. It also describes an experimental video-on-demand system being developed.

View full abstract

This paper presents a software system for performing amplitude modulation and demodulation in real time. This system uses a new method that approximates a tuning filter with a differential equation. By applying this method, the number of calculations for tuning can be reduced to one fourth as compared to a filter that uses convolutions, thus enabling high speed software modulation and demodulation.

View full abstract

One unsolved problem to the perfection of ATM networks is that of remotely controlled virtual path testing through the standardized Telecommunications Management Network (TMN) model. Thus, this paper addresses the two issues that arise when the TMN model is applied to virtual path testing. One issue is to define relevant information models. The other issue is to provide test resources with a concurrency control mechanism that guarantees a consistent test environment without causing deadlocks. To resolve these issues, alternatives to the concurrency control mechanism are shown and compared through computer simulations. Additionally, a method of defining information models is proposed. The proposed method ensures the easy storage and retrieval of intermediate test results as well as permitting the effective provision of concurrency control.

View full abstract