Multiscale image reconstruction for low bit-rate wavelet-based image coding

At low bit-rates, wavelet-based image coding demonstrates some advantages over most traditional block-based coding methods in that it does not introduce perceptually annoying blocking artifacts in coded images. However, the coded images still suffer various coding artifacts at edges and continuous regions, which are disturbing to human perception. A new post-processing method is proposed to improve the image quality for low bit-rate wavelet-based image coding using the technique of signal characterization from multiscale edges. First, we analyzed the edge degradation process based on an edge model. Then the edge degradation is characterized in terms of the distortions of modulus maxima of the dyadic wavelet transform. Through some experiments on a set of images, a group of compensation functions are developed for modulus maxima restoration. Thus, from the restored modulus maxima and with some appropriate constraints, we can reconstruct an image with improved visual quality and image fidelity (PSNR)