A DCT gain-shape vector quantizer for image coding

A discrete cosine-transform (DCT) image-coding technique using the gain-shape vector quantizer is presented. The original image is partitioned into many subblocks, and each subblock undergoes a two-dimensional DCT. The vectors to the quantizer are formed by selecting the same coefficients from th different transform subblocks. The normalized AC and DC vectors with mean removal can be modeled as standard distributions. Various level gain and shape codebooks are designed based on these models. The gain component is quantized by a scalar Lloyd´s quantizer (1982). The bit allocation for the vector quantizer is designed to minimize distortion. The technique has been applied to a real image to evaluate its performance.<>