Very low bitrate video compression based on GFA modeling

In this paper, a very low bitrate video coding scheme using the generalized finite automata (GFA) representation of the bitplane model of video sequences is proposed. Unlike the traditional motion estimate based MPEG scheme, where a group of pictures (GOP) in a video sequence is arranged into I, P and B frames, in the proposed scheme, all the frames in a GOP are first transformed into the wavelet coefficients and the coefficients in subbands are then modeled by the significant bitplanes. Finally, the bitplane model is represented by the genealogy tree structured GFA. The state indices and their transitions in the GFA representation are then entropy coded using the arithmetic code. Since the fractal similarities between the frames, wavelet levels and bit-planes are optimally explored by the GFA, the GFA model of the GOP is minimized, leading to a super low bitrate video coding scheme which can facilitate applications such as video streaming over low bandwidth networks, video streaming to mobile clients (PDA and cellular telephones), mobile-conferencing, video gaming, etc