An analysis based, perceptually lossless, texture compression method

In this contribution, a perceptually leaseless, model based texture coding method is presented. The model's parameters, consisting of a binary excitation image, the coefficients of an AR filter and the histogram of the original texture, allow to reconstruct a texture which is perceptually indistinguishable from the original one. The estimated binary excitation is compressed using a binary wavelet transformation specifically tailored to binary images, which allows a fast and lossless coding, followed by an entropy coding applied even to the other parameters. The proposed method leads to very good perceptual results superior to those of existing coders.