摘要： 针对Bandelets变换在图像去噪时产生图像边缘伪Gibbs现象的缺陷，提出一种利用Bayes收缩算法逐层估计噪声方差非下采样Bandelets域的图像去噪方法。实验结果表明，与基于Bandelets变换的去噪方法相比，该方法可以避免Bandelets变换中进行下采样而使图像不连续点处信息丢失导致的图像不稳定，较好地保持边缘细节，提高了峰值信噪比。

关键词: 多尺度几何分析, 非下采样Bandelet变换, Bayes收缩, 图像去噪

Abstract: According to the Gibbs effects of Bandelets transform denoising method, an image denoising method in non-subsampled Bandelets domain is proposed. It utilizes Bayes shrinking algorithm to estimate the noise mean square level by level. Experiment on image denoising shows that compared with Bandelets transform method, this method avoids information loss in discontinuous point and image non-stabilization caused by down sampling. It improves image edge preserving and peak signal-noise-ratio.

Key words: multiscale geometric analysis, non-subsampled Bandelet transform, Bayes shrinking, image denoising

中图分类号: 

• TN911.73