Image denoising based on non-subsampled shearlet transform

Abstract

In this paper, a novel method for image denoising based on non-subsampled shearlet transform (NSST) is proposed which adopts multi-scale geometry tool. Firstly, the image is decomposed into multi-scale and multi-direction sub-band images by using NSST. The NSST coefficients of each direction approach the generalized Gaussian distribution. We use the principal component analysis (PCA) for every similarity window of NSST coefficients. Then we use Generalized Gaussian model of non-local means method to handle the NSST coefficients. Finally, we reconstruct image with the new NSST coefficients to obtain the result. Numerical results show that our algorithm competes favourably with nonlocal means algorithms in the case of high noise.