Vector Quantization using Artificial Neural Network Models

Abstract

This paper describes our ongoing research into the construction of an Adaptive Vector Quantization (AVQ) image encode. We provide the motivations behind an AVQ encoder and report on our progress-to-date in realizing a Vector Quantization (VQ) encoder. We describe the hardware that has been built to compress video in real time using full-search vector quantization. This hardware implements a differential-vector-quantization (DVQ) algorithm which employs entropy-biased codebooks designed using an Artificial Neural Network (ANN). The theoretical properties of this codebook design method are discussed. We conclude with a description of the framework we are using to study various AVQ techniques with the goal of modifying our existing VQ hardware in order to realize an AVQ encoder.