基于差分曲率的非局部均值图像降噪算法

A Non-Local Mean Image Denoising Algorithm Based on Difference Curvature

传统非局部均值滤波算法忽略了图像块之间的结构相似性,造成搜寻相似像素不充分,使得一些图像细节被滤除.因此,利用差分曲率对图像边缘和斜坡等结构信息的良好检测性能,提出了一种基于差分曲率的非局部均值降噪算法.该方法充分利用图像块之间灰度值和差分曲率的欧氏距离共同确定权重,对图像块之间进行更好的相似性判断,进而优化传统非局部均值滤波算法.实验结果表明,与传统的非局部均值降噪算法相比,新算法能有效地保持图像边缘细节等信息,改善了传统非局部均值算法的去噪性能,取得了良好的降噪效果.

The traditional non-local mean denoising algorithm ignored the structural similarity of image blocks,resulting in searching similar pixels inadequately and making some image details filtered out.Combining with differential curvature that could detect the edges and ramps of the image well,a new non-local mean denoising algorithm based on difference curvature was proposed.Making full use of the Euclidean distances of the difference curvature and the gray value among the image blocks to jointly determined the weight,the proposed method obtained better similarity judgment to improve the traditional non-local mean filtering algorithm.Experimental results show that,compared with traditional non-local mean denoising algorithm,the new algorithm can effectively maintain the image edge details and other information,which improves the denoising performance of traditional non-local mean algorithm and achieves good denoising results.