关键词: 图像去噪, 非下采样Contourlet变换, 平移不变性, 空间自适应缩减, 各向异性窗口

Abstract: A new image denoising method via spatially adaptive Bayesian shrinkage in the Nonsubsampled Contourlet Transform （NSCT） domain was propsed in this paper. the generalized Gaussian distribution was utilized to model the statistics of individual NSCT domain detail subbands. Anisotropic elliptic windows were then constructed and applied to the description of the local contextual characteristics of each coefficient located at a certain detail subband. As a result, an NSCT domain image denosing approach with spatially adaptive Bayesian shrinkage mechanism was established. The experimental results demonstrate the effectiveness of the proposed denoising method. Compared with four shiftinvariant contourletrelated denoising methods, the proposed method provides preferable denoised results.

Key words: image denoising, Nonsubsampled Contourlet Transform (NSCT), shift invariance, spatially adaptive shrinkage, anisotropic window