A Comparison Of Growing And Pruning Balanced And Unbalanced Tree-structured Vector Quantizers
E.A. Riskin, Robert M. Gray · 2005
We examine the question of whether growing and optimally pruning balanced or unbalanced tree-strnctured vector quantizers will result in lower average distortion. It is shown that growing an unbalanced tree will not always lead to lower distortion than growing a balanced tree, even though the balanced tree has more structure. Conditions are presented under which it is guaranteed that pruning an unbalanced tree will always outperform pruning a balanced tree of the same initial average bit rate. Finally, we present a case where pruning a balanced tree may outperform, at some bit rates, pruning an unbalanced tree of lower initial distortion.