Adaptive prediction for lossless image compression

摘要

Lossless image compression is often performed through decorrelation, context modelling and entropy coding of the prediction error. This paper aims to identify the potential improvements to compression performance through improved decorrelation. Two adaptive prediction schemes are presented that aim to provide the highest possible decorrelation of the prediction error data. Consequently, complexity is overlooked and a high degree of adaptivity is sought. The adaptation of the respective predictor coefficients is based on training of the predictors in a local causal area adjacent to the pixel to be predicted. The causal nature of the training means no transmission overhead is required and also enables lossless coding of the images.The first scheme is an adaptive neural network, trained on the actual data being coded enabling continuous updates of the network weights. This results in a highly adaptive predictor, with localised optimisation based on stochastic gradient learning. Training for the second scheme is based on the recursive LMS (RLMS) algorithm incorporating feedback of the prediction error. In addition to the adaptive prediction, the results presented here also incorporate an arithmetic coding scheme, producing results which are better than CALIC.