基于改进非局部均值的红外图像混合噪声去除方法

Infrared image mixed noise removal method based on improved NL-means

传统的去噪算法无法有效去除红外图像中的条纹与随机混合噪声。针对这一问题,提出了一种改进的基于非局部均值(NL-means)的混合噪声去除方法。首先,分析了非局部均值算法处理混合噪声的问题,并用一组实验分析了红外图像块中混合噪声的特性。根据实验结果,文中用有色高斯模型对混合噪声进行建模,并基于Mahalanobis距离改进了传统的基于欧氏距离的块相似性度量方法,使之对图像中不同复杂程度的区域进行自适应。仿真和真实数据实验均表明:文中算法相比于传统的图像去噪算法,能较好地去除条纹与随机混合噪声。

Typical denoising algorithm were unable to effectively remove the mixed noise of stripe and random noise in infrared images.To solve this problem,an improved NL-means filter was proposed.Firstly,the problem in NL-means algorithm dealing with mixed noise was analyzed,and an experiment was performed to analyze the characteristic of the mixed noise in image patches of infrared images.Based on the experiment results,the mixed noise was modelled using colored Gaussian model,and the ordinary patch similarity index was improved using Mahalanobis distance instead of Euclidean distance,so that it adapts to the local with different complexity in the image.Both simulated and real data experiments show that the algorithm can effectively remove the stripe and random mixed noise compared with traditional image denoising algorithms.