Adaptive Shrinkage for Image Denoising Based on Contourlet Transform

We propose a new local adaptive shrinkage denoising approach based on the neighborhood characteristics of contourlet coefficients. Classical contourlet shrinkage denoising methods process the contourlet coefficients with a fixed threshold in each subband, without considering the clustering property of the coefficients. The shrinkage denoising method proposed in this paper determines the shrinkage threshold according to the neighboring contourlet coefficients, the scale of the coefficients and the noise level. Thus, the new threshold can preserve more significant coefficients that contain information of important singularity, and at the same time, attenuate more coefficients that contain information of noise when compared with the classical one. Our experiments show that the proposed shrinkage denoising method outperforms the classical contourlet shrinkage threshold method, in terms of both PSNR values and visual quality, especially for the images that include plentiful textures and edges.