基于双阈值引导滤波的单幅图像去雨算法  

Single Image Rain Removal Algorithm Basedon Double Thresholds Guided Image Filtering

作　　者：刘亚楠 曹风云[1,3] 王璐 LIU Ya-nan;CAO Feng-yun;WANG Lu(School of Computer,Hefei Normal College,Hefei 230601,China;Anhui Provincial Key Laboratory of Multimodal Cognitive Computation,Anhui University,Hefei 230601,China;School of Computer Science,Hefei University of Technology,Hefei 230009,China)

机构地区：[1]合肥师范学院计算机学院,安徽合肥230601 [2]多模态认知计算安徽省重点实验室(安徽大学),安徽合肥230601 [3]合肥工业大学计算机与信息学院,安徽合肥230009

出　　处：《兰州文理学院学报（自然科学版）》2022年第3期42-47,共6页Journal of Lanzhou University of Arts and Science(Natural Sciences)

基　　金：安徽省高等学校自然科学研究重点项目(KJ2020A0090,KJ2017A934);多模态认知计算安徽省重点实验室(安徽大学)开放课题(MMC202009);安徽省高校优秀青年人才支持计划(gxyq2017050)。

摘　　要：传统的基于图像分解的单幅图像去雨算法,容易存在雨痕残留、纹理信息丢失等问题.针对单幅图像去雨问题,提出一种双阈值引导滤波去雨算法.文中采用改进的引导滤波提取雨图的低频部分,由于高频部分是由雨图减去低频部分得到,在引导滤波中引入约束项,保证高频部分为非负图像.同时,利用计算过程中的方差值设置两个阈值,得到权重因子,在去除雨痕的同时保留纹理细节信息.对高频部分采用基于形态成分分析的图像分解算法得到非雨分量,低频部分与非雨分量相加,得到最终的去雨图像.实验结果表明所提算法可以较好地提升去雨质量并保留纹理信息.In the traditional single image rain removal algorithm based on image decomposition,the rain traces are easily to stay and the texture information is lost.Double thresholds guided filtering algorithm is proposed to remove rain streaks from a single image.The low-frequency part of the rain image is extracted by the improved guided image filtering.Since the high-frequency part is obtained by subtracting the low-frequency part from the rain image,a constraint is introduced in the guided image filtering to ensure that the high-frequency part is a non-negative image.At the same time,two thresholds are set by the variance value in the calculation process,and the weight factor is obtained.The texture information is retained while the rain mark is removed.For the high frequency part,the non-rain component is obtained by image decomposition algorithm based on morphological component analysis,and the low frequency part is added to the non-rain component to get the final rain-removed image.The experimental results show that the proposed algorithm can improve the quality of rain removal and retain texture information better.

关 键 词：去雨 引导滤波 图像分解 形态成分分析 阈值 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]