基于局部和非局部正则化的图像压缩感知

发布时间：2019-06-24 13:06

【摘要】：基于低秩正则化的非局部低秩约束(Nonlocal low-rank regularization,NLR)算法利用相似块的结构稀疏性,获得了目前最好的重构结果。但是它仅仅利用了图像的非局部信息,忽略了图像像素间的局部信息,不能有效地重建图像的边缘,同时Logdet函数不能很好地替代矩阵秩,因为它跟真实解之间存在着不可忽视的差距。因此,本文提出了一种基于局部和非局部正则化的压缩感知图像重建方法,同时考虑图像的非局部低秩性和图像像素的局部稀疏梯度性。选择利用Schatten-p范数来替代矩阵秩,同时选择交替方向乘子算法求解产生的非凸优化问题。实验结果表明,与传统的稀疏性先验重建算法和NLR算法相比,本文算法能够获得更高的图像重构质量。
[Abstract]:The nonlocal low rank constraint (Nonlocal low-rank regularization,NLR) algorithm based on low rank regularization makes use of the structural sparsity of similar blocks to obtain the best reconstruction results at present. However, it only makes use of the non-local information of the image, ignores the local information between the image pixels, and can not effectively reconstruct the edge of the image. At the same time, the Logdet function can not replace the rank of the matrix very well, because there is a gap between it and the real solution. Therefore, in this paper, a compression perceptual image reconstruction method based on local and non-local regularization is proposed, taking into account the non-local low rank of the image and the local sparse gradient of the image pixel. The Schatten-p norm is used to replace the rank of the matrix, and the alternating direction multiplier algorithm is selected to solve the nonconvex optimization problem. The experimental results show that compared with the traditional sparse prior reconstruction algorithm and NLR algorithm, the proposed algorithm can obtain higher image reconstruction quality.
【作者单位】：金陵科技学院计算机工程学院;南京晓庄学院信息工程学院;
【基金】：江苏省高等学校自然科学研究面上(16KJB520014,14KJB520012)资助项目江苏省社会安全图像与视频理解重点实验室(30916014107)资助项目国家自然科学基金(61375121)资助项目金陵科技学院博士启动资金资助项目
【分类号】：TP391.41

【相似文献】