Adaptive likelihood codebook reordering vector quantization for 1-D data sources

Abstract

This paper outlines an adaptive extension of likelihood codebook reordering (LCR) vector quantization. By providing a method for allowing the vector quantization to adapt in a predetermined way, the codebook may be adaptively reordered to allow more efficient encoding by giving preference to encountered vectors in the dictionary. In particular, adaptation allows the trained dictionaries to be more efficient in representing specific data. The difference in the training and testing sets produces different transition matrices which are used to encode testing vectors. The adaptive likelihood codebook reordering vector quantization adapts the a priori transition matrix obtained from training data set to the testing data set on an online instantaneous basis. This method yields improvements in coding rate when entropy coding is applied to the reordered indices obtained from the adaptive version of the LCR algorithm.