Color Image Denoising Using Sparse Representation Based on Separating Color Channels Locally

This article offers a novel approach to separate luminance and chrominance information that leads to better results in image denoising. Unlike usual methods, the luminance -chrominance channels created in this approach are not fixed channels that fit the dimensions of the original image but are three-dimensional arrays that are made up of groups of similar blocks in a local neighborhood. luminance -chrominance information of each group were extracted in all three RGB channels using PCA. In fact, using the PCA converter makes it possible to transfer similar parts to the luminance -chrominance space with coefficients specific to those areas, instead of transferring the entire image to the luminance -chrominance space with fixed and predetermined coefficients and then these channels participate in denoising operation directly. Image denoising in this approach, like the BM3D method, is based on two stages of initial and final reconstruction. So image denoising will be done by applying 3D wavelet transform and hard thresholding in the first stage and applying Wiener filter in the final stage. The innovation of this approach removes noise effectively. Moreover, it can improve the perception of luminance and chrominance information. Experimental results show that the proposed method can achive better performance compared to other methods such that the proposed technique improves the results by about 1dB in a low-noise situation and 3dB in a high-noise situation.