Performance analysis of multiresolution image denoising schemes

The denoising of natural images corrupted by noise is a long established problem in signal or image processing. Noise is signal dependent and is difficult to be removed without impairing image details. Multi-resolution methods are based on image transformations that reduce image resolution and they are able to segment objects of different sizes depending on the chosen resolution. In this work the transformations used for multiresolution analysis are Wavelet Transform (WTT), Contourlet Transform (CTT) and Non Subsampled Contourlet Transform (NSCT). One common approach found in the literature for image denoising involves manipulating the coefficients in the transform domain, example shrinkage followed by the inverse transform. In this work different shrinkage rules such as Universal shrink, VISU shrink, Minimax shrink SURE shrink Bayes shrink and Normal shrink are incorporated. These shrinkage rules combined with soft thresholding are applied to several Gaussian noise added test images. Experimental results show that NSCT outperforms WTT and CTT in terms of PSNR and preservation of edge information.