Mixed Noise Removal Method Based on Sparse Representation and Dictionary learning: WESNR

Noise removal is the fundamental problem in image processing.Knowledge of Noise Distribution is important in image denoising. Removing mixed noise from an image is since a difficult task as the characteristics of different types of noises are different.The commonly experienced mixed noise is impulse Noise(IN) together mixed with additive White Gaussian noise(AWGN).Various mixed noise removal methods are available but they are detection based methods. These methods first of all find the loacation of IN pixels and then remove the other noises. But these methods produce many artifacts when the density of mixed noise is high. Here, we introduce a very effective method named as weighted encoding with sparse nonlocal regularization (WESNR).In this method there is not impulse pixel detection and AWGN removal performs separately but it performs this both task in unified framework.In WESNR impulse pixel detection by weighted encoding is done to deal with IN and AWGN simultaneously.Nonlocal self-similarity and sparsity is combined in to a regularization term.WESNR achieves best mixed noise removal performance than any other methods.