基于噪声方差确定非线性扩散除噪声的最优停止时间

该文采用非线性扩散进行图像除噪声并在这个计算框架下提出利用噪声方差选择最优停止时间的方法。在利用非线性扩散进行图像除噪声时，每次迭代平滑掉的图像的方差大于平滑掉的噪声的方差时，迭代应该停止。为了在除噪声过程中正确地估计噪声的方差，该文构造一幅纯噪声图像跟实际的观测图像同步进行迭代计算，并把纯噪声图像的方差作为图像中噪声方差的估计值来辅助计算最优停止时间。针对非线性扩散的各项异性，提出了能够保持两种噪声同步变化的特殊的规整化项。新的规整化项在迭代纯粹噪声图像时使用，这样确保每次迭代都可以保持合成噪声与实际图像噪声的统计特性相一致。实验证明新的算法可以非常有效地选择合适的停止时间。

The Selection of Optimal Stopping Time Based on Synchronous Iteration of Noise and Image in Diffusion Image De-noise

This paper de-noises in image by diffusion filter. And a method of finding optimal stopping time is proposed. The criterion of stopping time is that the variance of image smoothed out is bigger than the variance of noise smoothed out. In order to estimate the variance of noise in iteration correctly, a pure synthesis noise as an image is synchronously iterated with the observation image in iteration, and the variance of pure noise image is taken as the estimation of the variance of noise in estimated image. According to the anisotropy of the regularization, a novel regularization term that can ensure the synchronous changing of the synthesis noise and real noise was proposed in this article. The new regularization term is put into use only in iteration of pure noise image, and the similarity of statistical properties between real noise and synthesis noise can be kept in iteration. Experiment confirms the effectiveness of proposed method to select optimal stopping time.