Block-Constrained Methods of Fixed-Rate, Entropy-Coded, Scalar Quantization

Motivated by the recent work of Laroia and Farvardin, this paper presents new reducedcomplexity methods for avoiding the bu ering problems associated with entropy-coded, scalar quantization. Basically, given a xed-size source block, these methods use dynamic programming and other techniques to search the sequences produceable by an entropycoded, scalar quantizer for one with minimum distortion subject to a constraint on the number bits produced by some binary encoding of these sequences. The result is that although some encoding methods might have a variable rate on the sample level, the overall quantizer has a xed rate on the block level. A general class of such methods, called block-constrained quantization, is introduced. A way to reduce the encoding complexity and several ways to to simplify the search complexity are found. A node-varying method with improved performance is given. New insight into the performance of block-constrained quantizers is presented. Compared to the original Laroia-Farvardin method, the results presented here show small improvements in performance and large reductions in complexity.