A perceptually lossless, model-based, texture compression technique

In natural scenes, still images as well as sequences, backgrounds, and obje cts' surfaces usually have a textural structure. Therefore, in order to eff iciently code images it is crucial to investigate the texture compression p roblem. In this paper, a perceptually lossless, synthesis-by-analysis textu re coding method is presented. The proposed approach is model based; the pa rameters of the model consist of a binary excitation signal and the parsimo nious representation of the reconstruction filter. The estimated parameters , which allow to synthesize, at the decoder site, a texture that is percept ually indistinguishable from the original one, are then compressed using a lossless strategy, which is based on a fast binary wavelet transformation s pecifically tailored to binary images. The proposed method leads to very go od perceptual results superior to those of existing techniques.