Classified vector quantization using fuzzy theory

Classified vector quantization (CVQ) is a technique for coding images that achieves good perceptual results while reducing the computational load of the process. A method that uses fuzzy theory to perform the classification involved in the CVQ technique is presented. To formulate the membership function of different categories, the concept of surface representation by triangular elements is introduced. This method yields a more natural classification scheme. The problem of determining the size of each subcodebook was also solved. The subcodebooks were built using fuzzy classifiers and the degree of accuracy of each codeword was tested by computing its fuzzy entropy. The codewords with larger fuzzy entropy were further split until an optimal point was reached.<>