基于时域自适应滤波及非局部平均的夜视图像去噪算法  

Night vision image denoising algorithm based on time domain adaptive filtering and non-local means

作　　者：刘小园[1] 衣扬 杨磊[3] Liu Xiaoyuan;Yi Yang;Yang Lei(Dept.of Electronic Information,Luoding Polytechnic,Luoding Guangdong 527200,China;School of Data ＆ Computer Science,Sun Yat-Sen University,Guangzhou 510275,China;College of Mathematics ＆ Informatics,South China Agricultural University,Guangzhou 510642,China)

机构地区：[1]罗定职业技术学院电子信息系,广东罗定527200 [2]中山大学数据科学与计算机学院,广州510275 [3]华南农业大学数学与信息学院,广州510642

出　　处：《计算机应用研究》2018年第6期1917-1920,共4页Application Research of Computers

基　　金：国家自然科学基金面上项目(61672546);广东省高等职业技术教育研究会2016年立项课题项目(GDGZ16Y079);广东省科技厅专项资助项目(2016A020212020)

摘　　要：为了增强夜视图像的可视性和清晰度,提出了一种适用于夜视环境的视频图像去噪和增强算法。首先在噪声消除方面,提出了一种基于卡尔曼结构更新的运动自适应时域滤波器;然后利用带有自适应限幅阈值的伽马校正调整RGB直方图来增加去噪视频图像的动态范围;最后,使用非局部平均(NLM)去噪滤波器消除剩余噪声。提出的方法能够直接用于颜色滤波阵列(CFA)原始视频图像,以便获得较低的内存消耗。在夜间环境(低于0.1 lx)下,利用200万像素CMOS传感器进行了具体测试,实验结果表明,相比现有的夜视图像增强方法,提出的视频图像增强法在夜间环境下均表现出了更加良好的性能指标。In order to enhance the visibility and definition of night vision image,this paper proposed a new algorithm of image denoising and enhancement. Firstly,this paper proposed a motion adaptive time domain filter based on Kalman structure updating. Then,it increased the dynamic range of the video image by adjusting the RGB histogram with the adaptive threshold. Finally,it used the non-local means（ NLM） denoising filter to eliminate the residual noise. The proposed method could be directly applied to color filter array（ CFA） of the original video image in order to obtain low memory consumption. In very dim light environment（ less than 0. 1 lx）,it carried out a specific test using a 2 mega pixel CMOS sensor. Experimental results show that compared to the existing low light-level image enhancement method,the proposed video image enhancement method shows better performance at night.

关 键 词：噪声降低 色调映射 非局部平均 夜视视频图像 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]