摘要
数字媒体视频图像大多通过不同的监控设备截图获取,为清楚获取其中的相关信息,提出了基于邻域一致性的数字媒体视频图像超清修复方法,以实现超清修复.通过数字媒体视频图像样本块和其邻域中已知块的非零相似度设置鲁棒结构稀疏度;依据邻域关联因子与鲁棒结构稀疏度,获取优先修复样本块;按照鲁棒结构稀疏度确定待修复样本块尺寸、邻域一致性约束权重与搜索区域窗口尺寸;按照最小误差平方和准则,在搜索区域内寻找最相似匹配块;按照邻域一致性约束权重,获取匹配块的稀疏表示信息,在优先修复样本块内填充稀疏表示信息,完成数字媒体视频图像超清修复.实验证明:该方法可以在保持图像结构部分连贯性的同时,超清修复不同程度破损的数字媒体视频图像;在不同数字媒体视频图像破损百分比时,超清修复后的数字媒体视频图像的峰值信噪比较高,噪声抑制均方误差较低.
Digital media video images are mostly obtained through screenshots of different monitoring equipment.In order to clearly obtain the relevant information,a digital media video image ultra clear repair method based on neighborhood consistency is proposed to realize ultra clear repair.The robust structural sparsity is set by the non-zero similarity between the sample block of digital media video image and the known block in its neighborhood;According to the neighborhood correlation factor and robust structural sparsity,the priority repair sample block is obtained;The size of the sample block to be repaired,the weight of the neighborhood consistency constraint and the window size of the search area are determined according to the robust structural sparsity;According to the minimum error sum of squares criterion,the most similar matching block is found in the search area;According to the neighborhood consistency constraint weight,the sparse representation information of the matching block is obtained,and the sparse representation information is filled in the priority repair sample block to complete the ultra clear repair of digital media video image.Experiments show that this method can repair damaged digital media video images in different degrees while maintaining the coherence of image structure;When the damage percentage of digital media video image is different,the peak signal-to-noise ratio of digital media video image after ultra clear repair is high and the mean square error of noise suppression is low.
作者
夏弘睿
赵静
XIA Hongrui;ZHAO Jing(Department of art and Design,Ma anshan Normal College,Ma anshan 243000,China)
出处
《太原师范学院学报(自然科学版)》
2022年第3期32-36,共5页
Journal of Taiyuan Normal University:Natural Science Edition
基金
安徽省2019年度高校优秀青年人才支持计划项目(gxyq2019193)
安徽省教学研究重点项目(2020jyxm1885).
关键词
邻域一致性
数字媒体
视频图像
超清修复
结构稀疏度
误差平方和
neighborhood consistency
digital media
video image
ultra clear repair
structural sparsity
sum of error squares