Image Denoising with Nonsubsampled Wavelet-Based Contourlet Transform

Contourlet has shown its charming performance in image processing. The multi-resolution and multi-direction characteristics ensure the transform a good ability in handling the abundant texture information in natural images. However, the contourlet is not shift-invariant because of the subsampled filter structure. Thus, the contourlet will cause the visual artifact in image denoising applications. In this paper, we construct a geometric image transform by combining 2D wavelet transform and nonsubsampled directional filter banks. The nonsubsampled filter banks enable the proposed method to be shift-invariance. The nonsubsampled directional filter banks are implemented via an i-level binary tree decomposition. In order to assess the applicability of the proposed method, we extend it to image denoising. The numerical results show that this method can obtain higher PSNR and restrain the visual artifact compared with using wavelet transform and contourlet transform.