IMAGE COMPRESSION USING BLOCK PATTERN-VECTOR QUANTIZATION

In this paper, a new coding scheme called the block pattern-vector quantization (BPVQ) is introduced. In BPVQ. each block pattern that is frequently encountered in real world images is encoded with a separate codebook designed specifically for that block pattern. BPVQ offers (1) lower search complexity than the ordinary VQ, at comparable rates, because each input image block is encoded with a small group of vectors that 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 reproduced edges are not staircased; and (3) very robust performance. If higher fidelity of the reproduced images is required, BPVQ can be easily extended to another technique called the high quality BPVQ (HQBPVQ). In HQBPVQ. each edge block is divided into four subblocks and each subblock 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 robustness than the BFVQ. A hybrid BPVQ/HQBPVQ approach provides excellent quality of the decoded images at rates of 0.7-0.8 bits/pixel.