In image coding using vector quantization, intravector correlation is successfully utilized for designing the codebook. As the number of dimension is made large, the performance of the vector quantization approaches to the rate-distortion bounds monotonically. However, from the practical viewpoint, the number of dimension cannot be made large enough to limit the intervector redundancy, and typically 16 or less, because the computational complexity grows as a exponential function of the number of dimension at the fixed coding rate. Therefore, much intervector redundancy cannot be removed by the vector quantization of small vector dimension, and should be utilized for achieving higher compression. From the above viewpoint, this paper proposes a new vector quantization scheme which utilizes intervector redundancy. In the proposed scheme, the codebook and its size are adaptively changed according to the local statistics extracted from a 32×32 pixel subblock by the clustering algorithm, and practically for every 32×32 pixel subblock the codewords of the codebook is selected from among the prepared set of codewords.
抄録全体を表示