Classified vector quantisation of images: codebook design algorithm

Classified vector quantisation (CVQ) of images is a vector quantisation-based coding method for preserving perceptual features while retaining simple vector quantiser distortion measures during codebook design and the encoding process. A new algorithm for CVQ codebook design, the ´classified nearest neighbour clustering´ (CNNC) algorithm, is presented. The CNNC algorithm is based on a classification process of small image blocks and on an agglomerative clustering algorithm, and is used to design simultaneously M codebooks for M different classes, defined for a CVQ system. The CNNC algorithm can be used with squared error and weighted squared error distortion measures employing one of two optimisation criteria which are presented and tested. In addition, a fast search algorithm is presented aimed at reducing computational efforts encountered during codebook design. The CNNC algorithm is shown to provide a systematic and effective method for CVQ codebook design making CVQ more feasible and easy to implement.<>