摘要
针对现有的超分辨率重建方法在复杂运动模式下的鲁棒性不强,以及容易出现边缘模糊和重影的问题,提出了一种基于时空非局部相似性的视频超分辨率重建算法(STSR),实现了视频视觉分辨率质量和细节清晰度的提升.该算法构建了一种快速鲁棒性的基于Zernike矩特征的视频帧间非局部模糊配准机制,并在此基础上通过时空域非局部相似性信息融合实现超分辨率重建.文章算法的优势主要在于不依赖于精确的亚像素运动估计,具有较好的旋转不变性及噪声鲁棒性,能够适用于局部运动和角度旋转等复杂的运动模式.实验结果表明,提出的算法不论是从主观视觉效果,还是从定量客观评价指标方面,均优于现有其他算法.
Aiming at solving the problem in the existing super-resolution methods, that is, the robustness is not strong under the complex motion patterns and the edge blur and ghosting artifacts are usually produced, this paper proposes a video super-resolution reconstruction algorithm based on spatio-temporal non-local siinilarity （STSR）, which implements the improvements for the video visual resolution quality and details clarity. We construct a fast and robust non-local fuzzy registration mechanism between video frames based on Zernike moment feature. And then the super-resolution reconstruction is implemented by fusion of the non-local similarity information in the spatio-temporal domain. The advantages of our algorithm mainly lie in the tact that it does not rely on accurate estimation of subpixel motion, and has higher rotation invarianee effectiveness and noise robustness, thus it can be adaptive to the complex motion patterns such as local motions, some angles of rotation, etc. Experimental results demonstrate that the proposed algorithm outperforms the other existing algorithms in terms of both subjective visual effects and quantitative objective evaluation index.
出处
《系统科学与数学》
CSCD
北大核心
2016年第9期1397-1409,共13页
Journal of Systems Science and Mathematical Sciences
基金
国家重点基础研究发展计划资助973项目2012CB821200(2012CB821206)
国家自然科学基金(61320106006,61532006,61502042)资助课题
关键词
视频超分辨率重建
时空非局部相似性
模糊配准机制
Zernike矩.
Video super-resolution reconstruction, spatio-temporal non-local simi- larity, fllzzy registration mechanism, Zernike moment.
