Optimal weight assignment in recursive vector quantization

Optimal weight assignment for recursive vector quantization (VQ) schemes is investigated. Recursive VQ, such as the predictive VQ and the finite-state VQ, manages to reduce the interblock dependence through effective prediction. Since the prediction is based on previously decoded blocks, the quantization error of the pixels used for estimation will feed forward to subsequent coding blocks. The quantization error of these estimating pixels can be reduced by appropriate weight assignment in the associated distortion measure. We investigate the weight assignment which optimizes the predictive mean-removed VQ. This result can be generalized to other recursive VQ schemes