Improvement of finite state vector quantization using classification

Abstract

Describes a new method for accelerating image compression by using a vector quantization (VQ) method. The authors propose a new, fast VQ technique called finite state classified VQ (FSCVQ) which is based on side match finite state VQ (FSVQ). FSVQ predicts the state of an input vector and selects a small codebook called the state codebook from the state. This state codebook reduces the number of patterns and enables faster calculation than conventional VQ methods. A distribution of pixel values in boundary areas in a block is used to select the state codebook because one can assume that the distribution of pixel values is almost the same between abutted boundary regions. In the authors' FSCVQ method, they added a classifier to FSVQ to reduce the calculation time and selected an appropriate state codebook. The authors also improved the S/N ratio of decoded images by using edge orientation to classify input vectors. In FSVQ, the boundary areas which are used to select the state codebook are the left and the upper side of a coding block. In contrast, the authors' method selects boundary areas according to the edge class of the coding block. Simulation results have shown that FSCVQ has a shorter calculation time than FSVQ because the state codebook of FSCVQ can predict a better state than can FSVQ. Moreover, FSCVQ yielded higher S/N ratios than FSVQ because FSCVQ preserved edges by using the classifier.