Image compression using texture modeling

We consider the problem of improving the performance of multiwavelets-based image coders through texture parametrization. Texture parametrization is designed to achieve higher compression rates while maintaining excellent visual image quality. Tradeoffs among these two conflicting goals, maximizing compression rate and minimizing distortion due to compression, are possible by taking into account the imperfections inherent in the human visual system (HVS). We present a statistical view of the texture parametrization using balanced multiwavelets and develop a hybrid image compression scheme. The statistical scheme leads to a new multiresolution-based texture parametrization relying on the accurate modeling of the marginal distribution of balanced multiwavelet coefficients using generalized Gaussian density (GGD). Furthermore, we show that the proposed texture parametrization scheme is computationally-efficient. The proposed hybrid codec can be seamlessly integrated in any embedded image coder while requiring minimal header data.