Multiple-codebook run-length vector quantization for image coding

It is observed that the utilization of spatial redundancy when using vector quantization (VQ) to code images improves the coding performance. One way of utilizing this redundancy is to segment an image based on textural tendencies, and to use multiple codebooks for each segment. The local persistence of each segment is the spatial redundancy, and this eliminates the need to indicate the type information with an index for every block. Furthermore, based on a joint optimization criterion, the optimum allocation of bits is achieved: according to this criterion, we may choose to keep the current codebook instead of switching to another one.