Block truncation coding using pattern fitting

摘要

Block truncation coding (BTC) divides an image into blocks having given size and then encodes each block by two representative gray levels and a bit-pattern. In this work a modified scheme of BTC is proposed where the computed representative gray levels are the bias and the contrast in each block. Secondly, instead of determining bit-pattern for each block, an optimum bit-pattern is selected from a pattern-book. Thus the index of the optimum pattern is used to encode in lieu of the explicit pattern. Thirdly, if the contrast is low the block is assumed to be smooth and bit-pattern is not required to reconstruct the block. This leads to significant reduction in bit-rate (bpp). Finally, the contrast component and the predictive residual of the bias component are entropy coded to achieve further reduction in bpp. Performance of the proposed scheme is measured in terms of peak-signal-to-noise ratio and bpp, and is compared with other recently reported methods.