Multiresolution tree structured vector quantization

In some applications of progressive image transmission, the images are viewed at several resolutions with fewer bits at lower resolutions. A multiresolution tree structured vector quantizer is developed to produce an embedded code, so that the quality of the image is optimized for the corresponding resolution at any number of bits. The resolution at which the image is viewed given a particular number of bits is determined by the specific decoder. The multiresolution tree structured vector quantizer presented generates the codebook by greedy tree growing, which is an extension of the generalized BFOS algorithm. The tree is grown one step further by splitting the node which will yield the best ratio of the change in distortion at the corresponding resolution of the current bit rate to the change in rate. The decoder has codewords of all resolutions obtained by optimal centroiding for a given resolution and a given encoder partition. The encoding of an image is essentially the same as the BFOS algorithm and the difference is that instead of having a fixed distortion measure, the distortion measure is defined for the corresponding resolution at a particular bit rate. This algorithm is compared with the generalized BFOS algorithm for image quality at different resolutions.