We previously reported a novel lossless image coding scheme using predictors and arithmetic code which were optimized for each image. The coding scheme adopts a new block-adaptive prediction method in which linear predictors are designed at the encoder side and sent to the decoder as side information. Since such a method utilizes image data which is not available to the decoder, it is categorized as forward type adaptive prediction. On the other hand, backward type adaptive prediction based on the least squares method has been also investigated for lossless image coding. Though this type of prediction requires no side information, it often suffers from instability of the prediction and/or increase of computational complexity. In our experiments, the forward type adaptive prediction generally provides better coding performance than the backward type, however, the latter one tends to have superiority especially in such a region as includes complicated edges. Thereupon, this paper proposes a new method which alternatively selects better prediction type in each block from the viewpoint of coding efficiency.
展开▼