An effective compression scheme for prediction errors on lossless image coding

In this paper, we propose a new effective lossless predictive image coding scheme. It intends to reduce the entropy of the prediction error images by an additional nonlinear processing. First, we use a block adaptive prediction scheme (BAP). The resulted prediction error images contain different statistics, block by block. The blocks in the active region contain large error samples. For an entire image, error values are distributed in Laplacian distribution. Then we select blocks containing many large error samples and halve all sample values of them, to reduce the distribution range to a half. By this way, the entropy of the modified prediction error image decreases. Of course, the address information of the halved blocks and the other related information are required as the side information. Experimental results show that the entropy of the modified prediction error image is reduced by 0.2057 bpp in average, compared with the first prediction error image, demonstrating the effectiveness of this scheme.