Codeword-dependent post-filtering for vector quantization-based compression

A postfiltering technique for vector quantization (VQ)-based image compression schemes is presented. This technique uses the codeword of the current and neighboring blocks to identify appropriate two-dimensional postfilters. The filters are designed using a minimum mean-squared error (MMSE)-based training procedure and are therefore optimal for the images in the training set. Simulation results show that the postfiltering technique yields significant improvements in visual quality and signal-to-noise ratio