点击次数：
论文类型：期刊论文
发表时间：2022-06-30
发表刊物：计算机辅助设计与图形学学报
期号：12
页面范围：2173-2181
ISSN号：1003-9775
摘要：Image deblurring is one of basic problems in the field of image processing and analysis. Since it is an ill-posed problem, a regularization is required to improve the stability of the solving process. In this paper, we propose a regularization method for motion deblurring based on local weighted total variation (LWTV) in terms of the local features of the image, and its corresponding solution based on alternating minimization. In the part of non-blind deconvolution, to overcome the shortcomings of traditional total variation (TV) method, we adopt the local variation of image as weights to increase the punishment on the flat area and reduce the punishment on the edge area. In the part of kernel estimation, we first extract the significant structures using relative total variation (RTV) method, then estimate initial kernel with the significant structure, and finally estimate the temporary image using LWTV model. In this way, the kernel can be obtained by alternating above three steps iteratively. Experimental results show that the proposed deblurring method can not only remove the blur and noise, but also keep the sharp edge and suppress ringing artifacts.
备注：新增回溯数据