A generic postprocessing technique for image compression

Abstract

A postprocessing technique is developed for image quality enhancement. In this method, a distortion-recovery model extracts multiresolution edge features from the decompressed image and uses these visual features as input to estimate the difference image between the original uncompressed image and the decompressed image. Coding distortions are compensated by adding the model output to the decompressed image. Unlike many existing postprocessing methods, which smooth blocking artifacts and are designed specifically for transform coding or vector quantization, the proposed technique is generic and can be applied to all of the main coding methods. Experimental results involving postprocessing four coding systems show that the proposed technique achieves significant improvements on the quality of reconstructed images, both in terms of the objective distortion measure and subjective visual assessment.