Fuzzy Joint Gaussian-Impulsive Noise Removal Using Joint Distribution Modelling in Sparse Domain

Abstract

Image denoising is trivial. It is considered that when multiple sources of noise act simultaneously such a task tends to be more critical. The distribution of resulting noise will possess irregular structure with heavy tail leading to fuzzy in detection and removal of noise from images. Most mixed noise removal schemes first detect the pixels with noise attack and then attempt to remove the noise. The proposed scheme is a single phase mechanism where the noise detection phase is absent. The proposed scheme uses sparse coding as a base and modifies the weight of the fidelity term so that the heavy tail of mixed noise distribution is approximated to Gaussian distribution. The simulation results prove the superiority of the proposed scheme using peak signal to noise ratio and feature similarity index. Results show that in the severe mixed noise case a PSNR improvement of 1% is achieved, whereas in the intermediate and little mixed noise cases a PSNR improvement of about 4% and 5% ae achieved.