Abstract: Spatially varying quantization schemes try to exploit the non-stationary nature of image subbands. One technique for spatially varying quantization is classification based on AC energy of blocks. Several different methods of subband classification have been proposed in the literature. One method is to optimally classify each subband and send the classification maps as side information. Although image subbands can be shown to be roughly uncorrelated, they are not independent. Naveen and Woods proposed a method in which classification is done based on the AC energy of the block corresponding to the same spatial location, but from the lower frequency band. In their method, inter-subband dependence is exploited to almost completely eliminate side information, albeit at the cost of decreasing classification gain. In this paper, we proposed a new method of classification based on vector quantization of AC energy n-tuples formed by energies of blocks which correspond to the same spatial location in the original image but belong to different subbands. This method allows us to reduce the side information at the same time maximizing classification gain for each band under the vector constraint. The performance of the new method is compared with the other two methods. The comparison is made based on conditional entropies as well as actual bit rates.!8
展开▼