A geometric vector quantizer coding technique is illustrated in thecontext of a full motion video coder based on a three-dimensional sub-bandframework. The original image data is decomposed into different spatial-temporalfrequency bands and based on the data in each band, different quantization strategiesare applied to the bands. The lowest spatial-temporal frequency band is coded usinga traditional three-dimensional switched predictor and optimum scaler quantizer.The non-dominant sub-bands are coded using the vector quantization approach toefficiently encode the images while appropriately exploiting the sparse, highlystructured nature of the data to design the codebooks. Variable resolution isachieved using these techniques and no training is needed to establish or maintainthe codebook. A fast codebook search technique is provided for identifying theoptimal codebook vector for representing a block of input data. Examples of twoand three level geometric vector quantizers are also provided.
展开▼