摘要
视频关键帧提取技术是对视频进行摘要来提高视频内容访问效率的一种操作。传统的方法主要采用聚类的方法,未给出可信的关键帧代表性描述。尝试基于图计算算法实现关键帧抽取,该算法可以将一段视频中候选帧及其之间的关系表示成一个相关图,通过各帧间基于相关性对相邻帧的分值分配进行迭代计算,实现候选帧内容代表性评价;并提出了一种高效的帧间相关性计算方法。该方法通过两帧图像的最大稳定颜色区域(maximally stable colour region,MSCR)的匹配情况判定它们的相关性。在测试视频上将该算法与传统算法进行了对比测试,测试的结果验证了该算法的有效性。
The keyframe extraction is a visual summary method. It enhances the accessibility to the visual content. Tra- ditional methods extract keyframes through clustering. These methods don＇t provide reliable descriptions of keyframe representative. This paper proposed a novel keyframe extraction method through a graph model representing the candi- date keyframes and the correlations between them. The representative of candidate keyframe was calculated through propagating grade between correlated candidate keyframes iteratively. To support the calculation of the representative, the paper introduced a correlation calculation method according to how well the maximally stable colour regions of two frames match to each other. The experiments were conducted on several test videos and the results validated our key- frame extraction method.
出处
《计算机科学》
CSCD
北大核心
2014年第8期286-288,315,共4页
Computer Science
基金
公安部重点研究计划项目(2011ZDYJGADX016)
北京高等学校青年英才计划项目(YETP1366)资助
关键词
关键帧提取
相关性计算
视频
Keyframe extraction, Correlation calculation, Video
