摘要
非局部变分采用关于目标像素点对称的区域作为寻找相似信息的目标图像块,会遗漏部分相似信息或无法找到相似信息,造成图像的非局部相似信息无法有效利用,图像复原性能有限。为更有效地利用图像非局部相似信息,提出以关于目标像素点非对称的区域作为目标图像块的非对称非局部变分模型。为有效求解该模型,通过图像平移将二维空间的非对称非局部变分模型转变为三维空间的对称非局部变分模型,并给出该模型的交替Bregman迭代求解过程。通过对比实验证明:提出的非对称非局部变分模型更好地利用了图像的非局部相似度信息,可复原更多的图像信息,获得的峰值信噪比与结构相似度都高于非局部变分模型。
Non-local total variation( NLTV) exploits a symmetrical target patch centered on target pixel in the similar patch search process,a lack of non-local similar patches is inevitable in some cases and image restoration performance is limited. In order to make full use of the non-local similarity of an image,a novel asymmetric patches non-local total variation( APNLTV) based on multi-shifted target patches is proposed. To optimize this problem,the proposed asymmetric non-local total variation is simplified to a symmetric non-local total variation in the 3-dimensional space through shifting the original image,then the Bregman iteration scheme is used to optimize the super-highdimensional non-local total variation. The comparing experimental results show that APNLTV is effective to utilize the non-local similarity,reconstruct more image information and gain higher peak signal to noise ratio and structural similarity index than NLTV model.
作者
陈明举
林国军
韩强
董林鹭
CHEN Mingju;LIN Guojun;HAN Qiang;DONG Linlu(Robot Technology Used for Special Environment Key Laboratory of Sichuan Province,Southwest University of Science and Technology,Mianyang 621000,China;School of Information Engineering,Sichuan University of Science&Engineering,Zigong 643000,China)
出处
《重庆理工大学学报（自然科学）》
CAS
北大核心
2020年第2期127-132,202,共7页
Journal of Chongqing University of Technology：Natural Science
基金
国家自然科学基金资助项目(41374130)
四川省科技厅项目(2018GZDZX0043,2019YJ0476,2019YJ0477)
