Multi-scale sparse denoising model based on non-separable wavelet

For the issue of image denoising, in order to avoid the traditional multi-scale sparse representation methods, which used blocks of different sizes as a base function to represent image, the non-separable wavelets were taken. Their advantages included revealing the multi-scale structure, depicting the texture structure under different scales, and separating different directions and different types of singularity structure in a certain extent. Based on non-separable wavelets, a multi-scale sparse denoising model in the wavelet domain was we established, and then a collaboration sparse model for the sub-bands contained similar structures was designed to enhance the stability and accuracy of the sparse representation. The results show that the denoising effect based on new approach is obvious superior to the K-SVD algorithm.