Adaptive VQ with Memory and Hierarchical Table LookupsEE 391

In this paper we present algorithms for performing adaptive vector quantization with memory. By using memory between adjacent blocks which are encoded, we can take advantage of the correlation between adjacent blocks of pixels to reduce redundancy. We use either nite state vector quantization or predictive vector quantization to provide the memory. To further improve performance by exploiting nonstationarities in the image, we use variable block sizes in the encoding. This is done by using a quadtree data structure to represent an encoding based on using variable block sizes. To reduce encoding complexity, hierarchical table lookup based schemes can be used to replace all the full search encoders. Also, in a another algorithm, we combine quadtree encoding along with wavelet transforms. In this algorithm, we simply do quadtree encoding individually on each of the subbands which result from a wavelet decomposition.