基于Zernike矩的运动图像序列时空超分辨率重建  

Motion image sequence space-time super resolution reconstruction based on Zernike-moment

作　　者：刘红刚[1] 杜军平[1] 梁美玉[1] 王超[1] 曹守鑫 

机构地区：[1]北京邮电大学计算机学院智能通信软件与多媒体北京市重点实验室,北京100876

出　　处：《中南大学学报（自然科学版）》2013年第S2期294-299,共6页Journal of Central South University:Science and Technology

基　　金：国家重点基础研究发展计划("973"计划)项目(2012CB821200;2012CB821206);国家自然科学基金资助项目(91024001;61070142);北京市自然科学基金资助项目(4111002)

摘　　要：提出一种基于模糊匹配机制的运动图像序列的时空超分辨率重建方法。采用光流方法对失真、模糊低分辨率图像进行运动补偿。通过相关性判断策略将像素分为相关像素和非相关像素,摒除非相关像素的干扰。对局部运动进行简单判断,对于简单运动的局部块使用灰度向量来表示局部特征,对于复杂的局部块计算Zernike矩来表征局部特征。实验结果表明:该方法能有效地补偿失真、模糊图像,可提高重建效果和匹配精度。A method of spatial super-resolution was proposed based on the fuzzy registration.Firstly,the compensated frame that is blurred was obtained using optical flow and motion compensation methods.Then,pixels were divided into two classes,related pixels and non-related pixels,through correlation judgment strategy.Moreover,the local patch motion was analyzed.For simple motion patch,gray vector was used to represent the local features.For complex motion patch,in order to improve the registration accuracy,the Zernike-moment was calculated to characterize the patch.Experimental results show that the proposed method can not only compensate the fuzzy frame,but also improve the performance of reconstruction.

关 键 词：模糊匹配 运动补偿 ZERNIKE矩 相关性分析 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]