Hadamard transform classifier for predictive classified vector quantization

Abstract

This paper describes a new vector quantization (VQ) technique called Predictive Classified Vector Quantization (PCVQ) which is based on the Classified Vector Quantization (CVQ) designed by Ramamurthi and Gersho. Whereas the classification information has to be transmitted to the decoder in the CVQ, it is predicted in PCVQ thereby reducing the bit rate while maintaining the same image quality as in CVQ. As class prediction depends on the classification process, classifier has to be accurately designed and tested. The predictor consists of an edge predictor and a mean predictor. Using the predicted edge and mean information, the class number can be generated in the decorder without it having to be transmitted.