Adaptive quantization for low bit rate video coding

We present a new adaptive quantization scheme in the framework of the visual telephony video coding standards (H.263/H.263+/H.263++). The scheme relies on having multiple sets of uniform quantizers for the lowest quantization levels. The decision about which quantizer to choose is based on the local rate distortion characteristics of the source block being quantized. Two schemes are presented: one which is within the syntax of H.263, and the other which requires a minor change in the syntax of H.263/H.263+. It is shown that up to about 10% savings in bit rates can be achieved by using the second method. With tuned VLC tables, the gains are even higher. Performance is further increased if a second degree of adaptivity is introduced by having multiple sets of VLC tables.

[1]  Kannan Ramchandran,et al.  Syntax-constrained encoder optimization using adaptive quantization thresholding for JPEG/MPEG coders , 1994, Proceedings of IEEE Data Compression Conference (DCC'94).

[2]  Philip A. Chou,et al.  Entropy-constrained vector quantization , 1989, IEEE Trans. Acoust. Speech Signal Process..

[3]  Michael W. Marcellin,et al.  Universal trellis coded quantization , 1995, Proceedings of 1995 IEEE International Symposium on Information Theory.