IMAGE COMPRESSION USING BLOCK PATTERN-VECTOR QUANTIZATION

In this paper, a new coding scheme called the block pattern-vector qua ntization (BPVQ) is introduced. In BPVQ. each block pattern that is fr equently encountered in real world images is encoded with a separate c odebook designed specifically for that block pattern. BPVQ offers (1) lower search complexity than the ordinary VQ, at comparable rates, bec ause each input image block is encoded with a small group of vectors t hat match its pattern; (2) very good edge reproduction, at rates of 0. 5-0.6 bits/pixel (bpp). to the extent that the majority of the reprodu ced edges are not staircased; and (3) very robust performance. If high er fidelity of the reproduced images is required, BPVQ can be easily e xtended to another technique called the high quality BPVQ (HQBPVQ). In HQBPVQ. each edge block is divided into four subblocks and each subbl ock is encoded with a codebook designed specifically for that subblock pattern. Impressive edge reproduction is obtained at rates of 0.95-1. 1 bits/pixel using the HQBPVQ with even lower complexity and more robu stness than the BFVQ. A hybrid BPVQ/HQBPVQ approach provides excellent quality of the decoded images at rates of 0.7-0.8 bits/pixel.