Complexity optimized vector quantization: a neural network approach

Abstract

The authors discuss a vector quantization strategy which jointly optimizes distortion errors and complexity costs. A maximum entropy estimation of the vector quantization cost function yields an optimal codebook size, the reference vectors and the assignment frequencies. They compare different complexity measures for the design of image compression algorithms which quantize wavelet decomposed images. An online version of complexity optimized vector quantization is implemented by an artificial neural network with winner-take-all connectivity. Their approach establishes a unifying framework for different quantization methods like K-means clustering and its fuzzy version, entropy constrained vector quantization or self-organizing topological maps and competitive neural networks.<>