基于二分K均值聚类和最近特征线的视频超分辨率重建方法  

Video Super Resolution Reconstruction Method Based on Bisecting K-means Clustering and Nearest Feature Line

作　　者：渠慎明[1,2] 王青博 刘珊 张东生[2] QU Shenming;WANG Qingbo;LIU Shan;ZHANG Dongsheng(Institute of Intelligent Network System, Henan University, Henan Kaifeng 475004, China;School of Software, Henan University, Henan Kaifeng 475004, China)

机构地区：[1]河南大学智能网络系统研究所,河南开封475004 [2]河南大学软件学院,河南开封475004

出　　处：《河南大学学报（自然科学版）》2018年第3期292-298,共7页Journal of Henan University:Natural Science

基　　金：国家自然科学基金项目(U1404618);河南省科技发展计划项目(172102210186;182102311066;162102210175;152300410195)

摘　　要：提出一种基于二分K均值聚类和最近特征线的视频超分辨率重建方法.该方法在生成图像块训练样本库阶段,为了解决传统K均值算法对图像块聚类时间复杂度较大的问题,设计一种基于二分K均值图像块聚类算法,用以实现快速构建初始训练样本库;在视频帧超分重建阶段,通过改进的最近特征线方法扩充样本库,并舍弃其中的不合适样本,提升样本库的表达能力.该视频超分辨率重建方法应用于基于关键帧的视频超分辨率重建,首先对时序上近邻的两个关键帧下采样并分块,然后使用图像块聚类算法构建初始训练样本库;对于这两个关键帧之间的低分辨率帧,使用改进的最近特征线方法扩充的训练样本库重建出对应的高分辨率帧.实验结果表明,提高了视频超分辨率重建的时间效率,同时也提升了每一帧重建图像的主客观质量,证明了算法的有效性.A video super-resolution（SR）reconstruction method based on bisecting K-means image block clustering and nearest feature line（NFL）was proposed.In the stage of generating image block training sample library,in order to solve the problem of large time complexity of image block clustering by traditional K-means algorithm,an image block clustering algorithm based on bisecting K-means was designed to realize the rapid construction of initial training sample library.In the video frame SR reconstruction phase,the sample library was expanded by the improved NFL method.The unsuitable samples were discarded,and the expressive ability of the sample library was improved.This method was applied to video SR reconstruction based on key frames.First,two key frames of the nearest neighbor on the time sequence were down-sampled and divided into blocks,and then an initial training sample database was constructed using an image block clustering algorithm.For the low resolution frames between the two key frames,the training sample library was extended to reconstruct the corresponding high-resolution frames by using the improved nearest feature line method.Experimental results show that the time efficiency of video SR reconstruction is improved,and the subjective and objective quality of each frame reconstruction image is enhanced,and the effectiveness of the algorithm is proved.

关 键 词：视频超分辨率 关键帧 二分K均值聚类 最近特征线 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]