Abstract: Due to the unreasonable problem of the weights assigned in the traditional non-local means algorithm, based on the Gaussian kernel, it combines a new cosine function to improve kernel function. Aiming at the problem of inaccurate judgments of the neighborhood similarity in this paper, similar function is introduced to measure the similarity of neighborhood grayscale matrices of the image, making the allocation algorithm in weight has been significantly improved. The experimental results show that, under different noise, improved non-local means algorithm significantly improves the de-noising performance.

Key words: image de-noising, non-local means, new cosine function, kernel function, neighborhood similar function

摘要： 针对传统非局部均值算法中权值分配不合理以及邻域之间相似性判断不准确的问题，利用新余弦函数与高斯核函数结合对核函数进行改进;引入邻域相似函数对图像邻域灰度矩阵间的相似性进行度量，使得算法中权值的分配问题得到明显改善。通过PSNR与直方图实验结果表明，在不同的噪声影响下，改进的非局部均值算法在去噪性能上有显著提升。

关键词: 图像去噪, 非局部均值, 新余弦函数, 核函数, 邻域相似函数