动态群稀疏约束场景知识建模的感兴趣监控目标超分辨率重建

项目名称： 动态群稀疏约束场景知识建模的感兴趣监控目标超分辨率重建

项目编号： No.61461032

项目类型： 地区科学基金项目

立项/批准年度： 2015

项目学科： 无线电电子学、电信技术

项目作者： 朱华生

作者单位： 南昌工程学院

项目金额： 43万元

中文摘要： 如何得到高分辨率的视频监控感兴趣目标图像是视频监控领域的热点与难点。稀疏模型是当前超分辨率重建的热点。基于流形一致性理论的稀疏约束超分辨率技术，并不适合实际监控图像的超分辨率重建。本项目以实际监控视频感兴趣目标的超分辨率重建为研究目标，探索基于动态群稀疏约束场景知识建模的感兴趣监控目标超分辨率重建理论与关键技术。首先，依据视频背景图像的稀疏表示一致性假设，建立基于动态群稀疏模型的视频监控图像场景提取模型，实现监控图像中场景和感兴趣目标的分离；其次，利用监控图像场景的近似性模拟实际的降质过程，针对实际监控图像高低分辨率稀疏系数流形不一致性问题，研究高低分辨率场景图像稀疏系数的映射关系和降质模型，建立动态群稀疏约束的半耦合结构化字典和降质映射函数学习模型；最后，根据半耦合结构化字典和降质函数，针对实际噪声的影响，研究基于主成份动态群稀疏和结构自相似的感兴趣目标超分辨率重构。

中文关键词： 退化图像复原；正则化方法；稀疏建模

英文摘要： Its a hot and difficult topic to gain super-resolution target of interest from images in surveillance research. Sparse modeling is the hotspot of super-resolution reconstruction. Sparse constraint super-resolution technique based on consistency of manifolds is not suitable for the actual monitoring. The goal of this study is to reconstruct target of interest in actual surveillance video, and explore the therory and Key technologies of super-resolution reconstruction of interest object based on dynamic group and sparse scenario knowledge model. Scene extraction modeling will be done firstly, according to consistency assumption of sparse representation of the background image in video, to separate the scene and the target of interest from surveillance images. Then the degradation process will be simulated via approximation of the scene of surveillance images. Manifold inconsistencies sparse coefficients for the actual monitoring of high and low resolution images, we will study the degraded model of sparse coefficient between them. A coupling semi-structured dictionary and degrade function learning model coupling sparse constraints and dynamic group will be established. Finally, we research the target super-resolution reconstruction based on the principal component of sparse and dynamic group interested in self-similar structure, according to the coupling semi-structured dictionary and degrade functions, considering the affect of actual noise.

英文关键词： Degraded image restoration;regularization method;sparse representation