Abstract: The selection of optimal stopping time for diffusion filtering was studied. It defined a correlation coefficient of de-noised image and noise by multiplying the image with a decreased function of gradient. Then, the selection of optimal stopping time equaled minimizing that correlation coefficient. This criterion can suppress the affection of minutiae to correlation coefficient and get a more accurate stopping time without knowing the noise variance. Simulation results show that, the stopping time determined by this criterion is near-optimal according to best PSNR, and the new criterion adapts to a wide range of noise levels.

Key words: mage de-nosing, diffusion filtering, stopping time, correlation coefficient

摘要： 讨论了图像去噪的扩散方程模型中所存在的最优停止时间选取问题。通过对图像点乘一个与梯度成反比的作用系数来构造一个新的去噪后的图像与去掉的噪声之间的相关系数,最优停止时间的选取就等价于极小化该相关系数。该准则能有效抑制图像中高频信息的存在对相关系数计算的影响,获得准确的最优停止时间,并无需预先知道图像中的噪声水平。仿真计算结果表明采用该准则所得的停止时间非常接近于基于最优峰值信噪比所得的停止时间,且广泛适用于各种噪声水平。

关键词: 图像去噪, 扩散去噪, 停止时间, 相关系数