庞彦伟, 周俊, 邓君坪, 何宇清
AuthorsHTML:庞彦伟, 周俊, 邓君坪, 何宇清
AuthorsListE:Pang Yanwei, Zhou Jun, Deng Junping, He Yuqing
AuthorsHTMLE:Pang Yanwei, Zhou Jun, Deng Junping, He Yuqing
Unit:天津大学电气自动化与信息工程学院，天津 300072
Unit_EngLish:School of Electrical and Information Engineering, Tianjin University, Tianjin 300072, China
Abstract_Chinese:针对单幅图像下, 基于稀疏表示的去雨算法存在残差较大而导致图像恢复效果不理想的问题, 提出了一种优化图像高频部分几何分量的去雨方法．首先采用平滑滤波做图像分解, 得到雨图像的高频部分; 然后结合稀疏表示与近邻传播算法分离出图像高频部分的雨分量, 用图像的高频部分减去雨分量并做平滑处理, 以此作为几何分量; 此外, 对稀疏表示过程得到的字典进行再分类, 完善雨分量与非雨分量的区分, 最后完成图像恢复．实验结果表明, 该方法能有效利用图像的几何信息来解决纹理恢复误差较大的问题, 实现更精确的纹理恢复和雨分量去除．
Abstract_English:To solve the problem of the unsatisfactory effect in single-image rain removal due to the large residual in sparse representation，an improved algorithm based on optimizing geometric components in high frequency part is proposed. First，smoothing filter is used to decompose image and obtain high-frequency part. Then，the rain component in high-frequency part is obtained via sparse representation and affinity propagation. The geometric component is acquired by subtracting the rain component from the high-frequency part，followed by a smoothing process. In addition，the dictionary got from sparse representation is re-classified to improve the classification between rain and non-rain. Finally，the restoration of non-rain image is completed. Experimental results show that the proposed method can effectively utilize the geometric information in the image to reduce the error in texture restoration and achieve high accuracy in texture recovery and rain removal.
Keyword_Chinese:单幅图像去雨; 纹理恢复; 稀疏表示; 字典分类
Keywords_English:single-image rain removal; texture recovery; sparse representation; dictionary classification

PDF全文下载地址:http://xbzrb.tju.edu.cn/#/digest?ArticleID=5811