非凸稀疏先验图像恢复建模理论和算法

项目名称： 非凸稀疏先验图像恢复建模理论和算法

项目编号： No.61271452

项目类型： 面上项目

立项/批准年度： 2013

项目学科： 无线电电子学、电信技术

项目作者： 卢成武

作者单位： 重庆文理学院

项目金额： 60万元

中文摘要： 稀疏表示和正则化理论的最新研究进展表明："非凸拟范数较正则化凸范数更能有效逼近'理想'稀疏性度量- - 零范数；Bregman距更能有效度量各类奇异正则化反问题的逼近误差"。这为图像恢复的稀疏冗余建模研究引入了新思路和新工具。本项目主要研究非凸稀疏先验下的图像恢复建模理论、数值算法和关键技术问题。主要内容：图像及其梯度稀疏性的非凸度量；非凸非光滑图像恢复的稀疏冗余建模、算法和重构误差的Bregman距度量；图像的结构化稀疏表示等。创新点：1. 利用稀疏性的非凸拟范数度量和恢复误差的Bregman距度量，探求更有效的图像恢复模型和快速算法；2. 引入非局部梯度稀疏性的非凸拟范度量和基于样例的正则化技术，探求更有效的保边保纹理感知重构模型；3. 拓展图像形态分量分析研究，引入各分量的非凸拟范数结构化稀疏度量和非局部特性，探求基于稀疏域投影的图像恢复建模。本课题预期在理论上有突破，方法和技术有创新。

中文关键词： 图像恢复；稀疏表示；广义全变差；低秩逼近；显著性检测

英文摘要： Advances in sparse representation and regularization research have shown that the nonconvex quasi-norm is more efficient than the regularization convex norm to approximate zero norm which is an ideal measurement for sparse. Bregman distance can efficiently estimate error for various singular regularization inverse problems. It introduces the new idea and tool for sparse and redundant representation modeling in image restoration. This project will investigate the nonconvex sparse prior modeling, numerical algorithms and key technologies .etc for image restoration. Main Contents: The nonconvex sparse measurement of image and its gradient;The sparse and redundant representation modeling,its algorithm and error analysis with Bregman distance for nonconvex nonsmooth image restoration;The structured sparse and redundant representation modeling for image. Innovations: 1. To explore more efficient models and algorithms for image restoration using the nonconvex quasi-norm to measure sparse and Bregman distance to estimate error; 2. By means of the sample-based regularization technology and the nonconvex quasi-norm to measure sparse of the nonlocal gradient, to explore more efficient sensing reconstruction models with preserving stucture and texture; 3. By introducing different nonconvex quasi-norm to characterize the s

英文关键词： image restoration；sparse representation；total generalized variation；low rank approximation；saliency detection