基于非下采样Contourlet变换的图像相关去噪

针对Contourlet变换不具有平移不变性,在图像去噪中容易产生伪吉布斯现象的情况,本文将具有平移不变性的非下采样Contourlet变换与相关去噪法相结合,采用不同的相关系数归一化方法,用Bayesian阈值代替传统的硬阈值来达到更好的去噪效果。实验表明,该方法提高了去噪后图像的峰值信噪比(PSNR),同时,有效保留了图像的纹理信息,避免了伪吉布斯现象,改善了图像的视觉效果。

Image correlation denoising based on nonsubsampled Contourlet transform

The Contourlet transform does not bear shift-invariance, and is liable to result in pseudo-Gibbs artifacts in image denoising. A method of nonsubsampled Contourlet transform, which bears shift invariance, is adopted combined with correlation denoising in this paper. Different correlative coefficient normalization methods are adopted to achieve better denoising effects by using Bayesian threshold denoising instead of conventional hard threshold denoising. The experiment results show that Peak Signal to Noise Ratio(PSNR) is increased in the image after denoising. The image texture information is preserved effectively, pseudo-Gibbs artifacts are avoided, and the visual effect of images is improved at the same time.