Adaptive Non-linear Predictor for Lossless Image Compression

Reference 1] Jaroslav Fojt k VV aclav Hlavv a c. Adaptive non-linear predictor for lossless image compression. Abstract. The paper proposes the new method for lossless image compression that performs wery well and results can be compared with other methods that we are aware of. We developed further the Slessinger's idea to represent an image as residuals of a special local predictor. The pre-dictor conngurations in a binary image are grouped into couples that diier in representative point only. Only residuals that correspond to the less frequent predictor from the couple is stored. An optimal predic-tor is based on the frequency of predictor connguration in the image. Two main extensions are proposed. (1) The method is generalized for grey-level image or images with even more degrees of freedom. (2) The method that works with addaptive estimator is proposed. The resulting FH-Adapt algorithm performs very well and results could be compared with most of the current algorithms for the same purpose. The predictor can learn automatically from the compressed image and cope even with complicated ne textures. We are able to estimate the innuence of the cells in a neighbourhood and use only signiicant ones for compression.