LVQ of image sequence source and ANS classification of finite state machine for high compression coding

Artificial neural system (ANS) classification has been applied to the total set of states of a finite-state machine operating as part of an image-sequence coder. It has been found that the classification of states allows the use of a much smaller number of representation states, thereby drastically reducing the storage requirements of the finite-state machine. The coder implements a scheme for high compression of teleconferencing image-sequence data. It utilizes neural-net-based learning vector quantization (LVQ) operating in the spatial domain on 16-dimensional vectors. The method is structured as a combination of an intraframe algorithm followed by an interframe algorithm, operating on a bundle of frames. The intraframe algorithm operates on the head frame of the bundle: the interframe algorithm follows in order to encode the remaining frames. Then, the encoding process repeats with a new bundle of frames. These algorithms are finite state-based. Simulation experiments have been carried out for a videoconferencing image sequence consisting of 20 frames of 112×96 pixels, with 25% average block motion. The representation vectors were of2×2×4 resolution. The results obtained have shown that for peak signal-to-noise ratio (PSNR)=32 dB, the required bit rate is 0.08 to 0.10 b/pixel