Recently, there are many apprications using neural network model which process parallel distributed information and have similarity to human information processing. Neural network models for vision are very significant for some application of image communication and other picture processing. This article discusses a neural netowrk model which reduces redundances of data with prediction based on cybernetics of Wiener. This model selforganizes like neurons of retina which is input part of human vision, when input data are real picture.
Applying our newly developed human vision system model to compound sinusoidal patterns, we obtained following results; (1) The calculated optimum image observing states explain qualitatively the experimental human vision responses to dithered images. (2) The experimental contrast threshold characteristics to compound gratings can be quantitatively simulated by using our vision model.
Temporal characteristics of visual motion processes were examined in terms of spatial combination of motion signals by using random-dot stimuli. Several random-dot fields were displayed successively. Perceptual motion directions of a central terget region were measured by varing duration and numbers of the field displayed. The results showed that the motion detection mechanisms having a surround inhibition property were also found to have a temporal selectivity. The model of this mechanism was discussed. The effects of temporal integration of the motion signals in the spatial combination of those were also shown and discussed.
Fundamental knowledge is surveyed concerning representative methods for 3D display, and psychological and physiological properties in stereoscopic vision. Then some recent activities in our laboratories are described. Those are flicker-free field-sequential 3D television system for experiment, experimental results on binocular disparities and fusion range, the temporal characteristics, binocular disparity modulation transfer functions, etc.. Concerning activities on 3D system and display in foreign countries, researches in two organizations in Germany. HHI and IRT, are introduced. They are conducting the researches with methods sharply contrasted. Some problems in 3D pictures are also described.
We have obtained digital I, Q filters which are used in NTSC digital encoder. The proposed I and Q filters have the taps of 30 and 15 respectively, and satisfy both the frequency characteristics of the NTSC signal specification and smooth step responses.
This paper describes a 64-kb/s full-color video CODEC using DSPs. The coding algorithm is mostly based on CCITT Reference Model 5,which uses motion-compensated interframe prediction and DCT. The image format is 1/4 CSIF. The coder, as well as the decoder, consists of a personal computer. DSP boards, and other boards. Each DSP board has two DSP circuits, and many DSP boards can work concurrently. Experimental results show that this CODEC can transmit 1.65 frames/sec when three DSPs are used at each end, and if nine DSPs are used, the theoretically expected performance is 4.46 frames/sec.
This paper discribes a high efficiency HDTV CODEC developed for digital transmission at 100Mbit/s, which corresponds to the fourth level bit rate in the Japanese digital hierarchy. This CODEC employes a spatio-temporal adaptive filter for noise reduction and intrafield/interframe coding algorithm using extrapolative/interpolative prediction. Some parallel processing techniques were adopted for hardware implementation. From a result of digital transmission experiments using optical fiber, it was clarified that 100Mb/s HDTV transmission could be realized without any picture degradation.