多尺度自适应配准的视频超分辨率算法  

Video Super-Resolution Algorithm Via multi-Scale Adaptive Block-Matching Registration

作　　者：卢涛[1,2] 章瑾[1,2] 陈白帆[3] 管英杰[1,2] 

机构地区：[1]武汉工程大学计算机科学与工程学院,湖北武汉430205 [2]智能机器人湖北省重点实验室(武汉工程大学）,湖北武汉430205 [3]中南大学信息科学与工程学院,湖南长沙410083

出　　处：《武汉工程大学学报》2016年第2期178-184,194,共8页Journal of Wuhan Institute of Technology

基　　金：863计划项目(2013AA12A202);国家自然科学基金项目(61172173),(61502354);湖北省自然科学基金项目(2014CFA130),(2015CFB451);湖北省教育厅重点科研项目(D20141505);湖南省科技计划项目(2014FJ3157)

摘　　要：为了提升制约视频超分辨率重建质量的多帧配准精度,提出了一种基于图像块多尺度自适应配准的视频超分辨率算法.依据图像帧的块内容自适应选择配准尺度,运动边缘信息采用高精度配准,然后将运动向量补偿到多帧图像超分辨率重建代价函数中,利用最大后验概率算法迭代优化高分辨率视频帧.仿真表明:多尺度自适应配准算法不仅提高了配准精度,还提升了视频超分辨率重建图像帧的主客观质量,证明了多尺度自适应配准在视频超分辨率重建中的有效性.To improve the registration accuracy which influences the performance of video super- resolution algorithms,we proposed a video super- resolution algorithm based on multi- scale adaptive block- matching registration. Registration scale was selected with block contents and the moving edge information of video was represented in high registration accuracy. Then,the registration motion vectors were compensated into the cost function of multi-frame super-resolution,and the high-resolution outputs frames were iteratively optimized by the algorithm of maximum a posteriori estimation. Simulation results show that the adaptive multi- scale registration not only enhances the accuracy of video registration,but also improves the performance of video super-resolution,which demonstrates the validity of adaptive multi-scale registration in video super-resolution reconstruction.

关 键 词：视频超分辨率 多尺度块匹配 自适应配准 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]