基于非下采样剪切波变换域三变量模型图像去噪算法

Image Denoising Using a Trivariate Model in the Nonsubsampled Shearlet Transform Domain

结合非下采样剪切波变换域三变量阈值滤波和多分辨引导滤波,本文提出一种去除高斯白噪声的图像去噪的有效方法。在非下采样剪切波变换域中,以三变量非高斯模型对方向带通子带系数间相关性进行建模,采用最大后验估计理论推导出三变量收缩阈值函数。此外,对低频子带系数采用多分辨引导滤波进行平滑处理,以达到更好的噪声抑制效果。实验结果显示,本文所提去噪方法可以有效抑制噪声同时保留更多的图像细节信息,与其他滤波算法相比,该去噪算法可得到更高的客观数据及更好的视觉效果。

We present an efficient algorithm for removing white Gaussian noise from corrupted images by incorporating a nonsubsampled Shearlet transform （NSST）-based trivariate shrinkage filter into a multiresolution guide filter. In the NSST domain, coefficients are modeled as a trivariate Gaussian distribution, accounting for the statistical dependencies among interscale and intrascale transform coefficients. A nonlinear trivariate shrinkage function is derived using a maximum a posteriori （MAP） estimator. To obtain better denoising results, low-frequency sub-bands are smoothed using a multiresolution guide filter. Experimental results show that our algorithm is very effective in eliminating image noise, and performs better than other denoising techniques.