This paper describes the automatic car recognition system which extracts the car position from trafic images and classifies it into several car types. As the characteristic features of car types appears in its silhouette, this system has been equipped with two cameras ; one of which picks up the side image of the car for the classification of the car type and the other one picks up the upper left side image for the detection of 3D positions. Processing these two images by computer, the system can recognize the car types and 3D position of them.

View full abstract

From a good man-machine interface view point, computer recognition of human actions seems to be an effective method for data input. Hand shape recognition is a typical example of this method. In this case, non-contact recognition method is preferable. This paper describes hand recognition method with the use of image processing technique. In this method, 3D feature points of hand shapes are extracted by 2 TV cameras. Then the hand shapes can be recognized by comparing with the hand model data in the computer. This method seems to be applicable to wide variety of hand positions and to provide a high recognition in the computer.

View full abstract

A lattice filter has a various and excellent characteristics. Previously, the authors presented a design method of three-dimensional (3-D) lattice filter. However, the condition for cheking the stability of this filter was not derived. In this paper, we show the sufficient lattice parameter stability conditions for the 3-D AR lattice filter of the case of cubic-stace, and then as a tractical application of the proposed synthesis lattice, the all-pole spectral modeling is derived.

View full abstract

The design of two dimensional (2-D) recursive digital filters is currently receiving wide attention. These filters have many applications, including seismic prospecting and geophysics. However, the autoregressive (AR) lattice parameter approach, in particular has met with much less success. In this paper, we present a new design algorithm of the asymmetric half-plane model, which is adopted to achieve full design flexibility, and show the sufficient lattice parameter stability conditions for this filter. Furthermore, the design procedure is used to calculate the lattice parameter from the prescribed frequency characteristics. Then, using digital computer simulations, the effectiveness of the proposed filter is evaluated.

View full abstract

In previous paper, the authors proposed a design method of two-dimensional lattice filter using quater-plane support. However, this filter is not able to apply to ARMA model. The purpose of this paper is to construct a new two-dimensional filter with generalized lattice structure, by replacing a delay element to an all-pass circuit. Also, an adaptive algorithm based on LMS algorithm isconstructed and then the effectiveness of the proposed method is evaluated using digital simulations.

View full abstract

In computed tomography, the noise contained in projection will significantly affect the reconstruction. In this paper we proposed to use an adaptive windowing method to filter the noise of the projections, followed by the use of conventional algorithm. The proposed adaptive windowing method can filter the noise without any blurring of the edge. The applicability of the proposed method for CT has been demonstrated by computer simulations.

View full abstract

For effective neutron imaging with penumbral aperture, the aperture must be made from a thick substrate in order to provide a sufficient contrast image. However, it is difficult to obtain a isoplanatic point spread function for the thick aperture. The straightforward image reconstruction such as Wiener filter will introduce some significant distortion for a large object because of non-isoplanaticity. In this paper we proposed to use genetic algorithms (GA) for reconstruction of neutron image. Since in GA, we don't use the inverse of the point spread function, it is possible to obtain a distortion-free reconstruction even for a large object.

View full abstract

Computer images have permeated many aspects of modern society such as commercs, education and entertainment. Computer animation is typical of artificial image sequences. A great number of image sequence data need to be compressed so as to be transmitted in limited bandwidth or stored in finite capacity memory. Video compression standards are not well suitable for artificial images created by computers. This paper proposes to use structure runlength coding to compress animation sequences. However, extending simply the original structure runlength coding in image sequence may lead to that compression ratios become bad at the case of higher spatial correlation than temporal correlation. This paper suggests to increase a more symbol standing for temporal relation so as to improve the basic structure runlength coding. How to use the new symbol results in three algorithms. Experimental results show that these new algorithms are superior to original structure runlength coding.

View full abstract

Motion compensating technique is one of the most important technique for standards conversion and image coding. It is difficult to estimate accurate motion vectors at covered and uncovered background and a boundary of moving object. This paper proposes a new algorithm of motion detection using motion trajectory and a new assignment filter using the luminance intensity of moving area. This paper describes the computer simulation results.

View full abstract

This paper describes an efficent motion compensation (MC) method using edge adaptive motion vector interpolation. In the proposed method, interpolation is adaptively performed according to the presence of a segment boundary between control grids. The segment shape of the current frame is predicted from segmentaion results of the reference frame and the control grid motion vectors for the current frame. Here, segment shape information does not need to be coded, since the reference frame is the local decoded picture. Simulation results show that edge adaptive motion vector interpolation significantly improves the prediction performance especially at segment boundary. The SNR of the coded sequence obtained using the proposed MC method is about 1.7dB higher than that using the conventional block-based one.

View full abstract

In this paper, reversible transform coding systems of images are proposed. Because conventional transform coding requires very great numbers of levels of transform coefficients to reconstruct lossless images, data compression is not realized. Therefore we propose some reversible transform coding systems which don't require very great numbers of levels of transform coefficients. It is shown that the reversible subband coding system which was proposed by us ago is expressed as overlap transform coding. Simulation shows that the proposed systems have almost the same performance as DPCM and overlap transform coding has the best performance. We also investigate the possibility of progressive image display.

View full abstract

As the tool for creating the multimedia contents for various applications, the demand for the encoder is increasing. We had developed an experimental real-time encoder conforming to MPEG2. This encoder includes three main functional blocks of MPEG standard, video encoder, audio encoder and system multimplexer. As a result of processor based implementation of system multimplexer, the encoder could output two kinds of MPEG-2 stream, Program Stream and Transport Stream. By using the MPEG-2 decoder LSI for local decoder block and providing the SCSI-2 bus as the host PC interface, the encoder enables a compact and flexible realization of MPEG-2 encoding system.

View full abstract

A 45 Mbps HDTV digital codec is described, aiming at transmission of HDTV signals via conventional standard television networks. In order to achieve satisfactory picture quality for contribution and primary distribution applications, new advanced coding techniques of a mixed mode coding, 3 dimensional structure quantization and 6 segments parallel coding are introduced into a motion compensated interframe and intrafield adaptive DCT coding. From experiments using a practical hardware codec, it is cleared that this codec provides significants better coding performance than MPEG-2.

View full abstract

Cell loss resilience has been main concern of video transmission through ATM networks, and bit error resilience has been rarely studied. Recently, the degradation of the transmitted video quality due to bit errors has been noticed. In this paper, using HDTV CODECs for ATM networks we had developed, we investigated the effects of bit errors and the combination of bit errors and cell loss to the transmitted video quality and made sure the effectiveness of the error resilience schemes to bit errors, which we had already exploited originaly for cell loss. The effects of bit errors and cell loss to audio transmission through ATM networks were also exhibited.

View full abstract