Adaptive Entropy-Coded Subband Coding of Image Sequences

We describe a new approach to image sequence coding based upon variable-rate entropy-constrained subband coding (ECSBC) and, furthermore, develop the corresponding practical implementation of the ECSBC scheme for fixed-rate channels by extending recently developed adaptive entropy-coded quantization (AEC) techniques. Although the entropy-constrained design of subband coding systems provides improved coding efficiency compared to level-constrained design approaches, the resulting coding system generates variable-rate outputs which must be buffered before fixed-rate transmission. In this case, the finite buffer, however large, that interfaces the encoder and the channel will eventually overflow or underflow resulting in a catastrophic loss of encoder-decoder synchronism with an associated large amount of distortion. A buffer-adaptive arithmetic-coded implementation of the ECSBC scheme, called adaptive entropy-coded subband coding (ECSBC/AEC), is described to completely eliminate the associated encoder buffer overflow/underflow problems, even with a very small encoder buffer. We demonstrate that the ECSBC/AEC scheme operating on real-world image sequences performs very close to the original ECSBC scheme whose performance is achievable only with an encoder buffer of infinite size. Moreover, promising results on the low-rate interframe coding of real-world monochrome and color image sequences are presented using the ECSBC/AEC scheme combined with a hierarchical motion-compensated prediction method.