We consider a multipoint teleconference system without MCU in LAN environment. While video data must be broadcasted to all terminals in a multipoint system, its throughput is affected by the number of connected terminals or the processing capability of each terminal. In this paper, we propose to transmit video data via several shared momeries which can be accessed by all the terminals in order to implement a scalable system to the throughput. By using the shared memories, each receiving terminal can skip video frames by itself according to its own capability. Moreover, conference space information including eye direction of each participant is considered as a supplementary information which is useful for a multipoint system.

View full abstract

In this paper, We describe a video storage system using a video communication system for LAN (PMTC/LAN) with UNIX workstation. The workstation emulates the PMTC/LAN'S real-time communication protocol on TCP/IP. Audio and Video packets are stored by the workstation in real-time, and are trasmitted to PMTC/LAN for playback. Real-time communication capabilities of the workstation was evaluated as a prototype of a video server system.

View full abstract

Prototype model of Hi-vision CODEC(HERCULES-32), that works at the channel speed of 32 Mbps, has been developed for the purpuse of monitering electric power facility, construction site, etc. It employs DCT, inter-field motion compensation and other techniques for compression coding. Stereo sound, data transmisson for camera control. NTSC compression coding and NTSC/Hi-vision conversion are some of its functionarities. This paper descrives an outline of the system.

View full abstract

In This paper, adaptive temporal prediction methods are studied to improve coding performance of the motion-compensated prediction-discrete cosine transform(MC-DCT) coding. The coefficients of adaptive prediction are designed with least squares method In the cases of coding predictor coefficients as side information and estimating them from previously reconstructed signals in the neighbor of DCT coding block. Simulation results show that these methods improve coding performances about 0.3〜2.0 dB compared with conventional methods using loop filter which is recommended in CCITT and using an adaptive loop filter which is designed to be edge preserving noise reduction filter.

View full abstract

This paper deals with subband coding for image compression. In this paper, parallel all-pass filters are used as the subband filters. We first propose an extension method at the boundaries of images, and show its superiority in computational efficiency and coding gain against the conventional periodic extension method. Next, we show that spatial locallzation of analysis filter is important as well as frequency localization, and propose the method of using different filter coefficient for each layer in binary tree coding. This method ensures coding gain of approximately 0.7dB for genelal half tone images.

View full abstract

Subband coding with non-orthogonal filter bank involves allocating the proper frequency characteristics for each band. In this manner among many possible combinations, one have the best coding gain due to better power compaction For instance, it has been experimentally shown that in SSK filter bank. 5taps lowpass and 3taps highpass analysis filters are much better than the reverse combination. However in this paper we will show that it is not true for all the signals especially in the intermediate bands. By means of this fact we propose an adaptive method which always selects the better filter bank before analysis.

View full abstract

The existing block-matching algorithm detects a pair of image blocks with some local similarity between two image frames. The algorithm works well under the condition that an image block is contained in a single image area which has a single optical flow. However, in such a case that an image block is defined near the boundaries between a still area and a moving area, the block containes multiple image areas with different optical flows. The existing block-matching algorithm does not work well for such blocks. In this paper, we propose a block-matching method based on a novel distance criteria, which can be applied to such image blocks. If necessary, our novel block-matching method will devide one image block into multiple areas and assign different optical flows to the areas. This paper presents an algorithm for the novel block-matching method and demonstrates its good performance of motion estimation by computer simulations.

View full abstract

Block noise occurs at the low bit rate coding when orthogonal transform coding such as DCT is used for high efficient redundant reduction. The block noise is mainly due to quantization errors in DCT coefficients and discontinuities in luminance level at block boundaries of adjacent blocks. We propose a postprocessing method to reduce the block noise of decoded image. An adaptive filtering is performed for mean values of subblocks divided into MxN with thresholds varied with the inside area and the boundary of DCT blocks. The simulation results show that the picture quality is improved by reducing block noise without blurring.

View full abstract