Applications of variable rate vector quantization to speech and image coding

Eve A. Riskin, Tom Lookabaugh, Philip A. Chou, Robert M. Gray

Research output: Contribution to conferencePaperpeer-review

Abstract

Summary form only given, as follows. A recently developed technique for variable-rate vector quantizer (VQ) design has been applied to both speech and image waveforms. The design algorithm is an extension of Breiman, Friedman, Olshen, and Stone's algorithm for optimal pruning in tree-structured classification and regression. The algorithm finds subtrees of a given tree-structured VQ, each one optimal in that it has the lowest average distortion of all subtrees with the same or lesser average rate. Since the resulting subtrees have variable depth, natural variable-rate coders result. Application of the design algorithm to waveform coding of speech at 0.75 b/sample using the mean-squared error distortion measure led to gains over ordinary fixed-rate full-search VQ of up to 1.8 dB. The algorithm was also tested on a medical image coding application using a data set of 25 magnetic resonance image (MRI) brain scans at 8 b/pixel. For compressed images at rates of .1875 b/pixel, the improvement in the SNR for the pruned images over full-search VQ was 2.47 dB. At a higher rate of .4375 b/pixel, the gain was 1.39 dB over full-search VQ. More importantly, both of the pruned images had less blockiness and better edge reproduction than the full-search images. The authors expect applications of this algorithm at higher rates to provide images of adequate quality for medical diagnosis.

Original languageEnglish
Pages224-225
Number of pages2
StatePublished - 1988

Fingerprint

Dive into the research topics of 'Applications of variable rate vector quantization to speech and image coding'. Together they form a unique fingerprint.

Cite this