基于小波预处理及自适应权函数的非局部均值滤波算法

Non-local Means Algorithm Based on Adaptive Kernel Function and Wavelet Threshold Preprocessing

非局部均值滤波算法因其良好的去噪效果受到了广泛关注,内容主要集中在算法加速,搜索框、相似框尺寸以及平滑参数的自适应设置等方面.然而,研究发现,在非局部均值滤波过程中,欧氏距离权函数对去噪效果也有较大影响.本文通过对权函数的分析,根据图像含噪情况、像素点的局部结构,自适应选择权函数,提高了去噪效果.此外,由于非局部均值滤波在噪声强度大时去噪效果不佳,而小波变换能够很好区分图像高频信号与噪声,本文先采用小波阈值去噪方法对图像进行预处理,再进行自适应权函数非局部均值滤波.仿真实验表明,本文方法在去噪效果上有明显提升,尤其适用于噪声强度较大的场景.

The non-local means algorithm（NLM）has received wide attention due to its good denoising effect and the content is focused on acceleration of algorithm、adaptive search box、patch size and smooth parameter. However,it is found that the kernel function used in calculating Euclidean distance has a great influence on the denoising effect. According to the image＇s noise level and the local structure of each pixel,the kernel function is analyzed and adaptive kernel function can be selected to improve the denoising effect. Furthermore,the performance of NLM is not good at high noise level and Wavelet transform can distinguish the high frequency signal and the noise fine. In this paper,the wavelet threshold denoising algorithm is used to preprocess the image,and then use the NLM with adaptive kernel function. The simulation results show that our method has a significant improvement in the denoising effect,especially in the scene of the noise intensity.