一种重复视频的快速检测算法  

A Fast Algorithm for Near-duplicate Video Detection

作　　者：刘大伟[1,2] 朱明[1,2] 

机构地区：[1]中国科学院网络传播系统与控制重点实验室,合肥230027 [2]中国科学技术大学自动化系,合肥230027

出　　处：《小型微型计算机系统》2013年第6期1400-1404,共5页Journal of Chinese Computer Systems

基　　金：国家科技支撑计划项目(2011BAH11B01)资助;国家自然科学基金项目(60975045)资助

摘　　要：随着网络多媒体应用的增加,各种网络视频应用需求大规模重复视频检测的方法,尤其对检测的快速和有效性要求逐渐增高.提出一种基于多层视频内容分析的快速有效检测重复视频的算法.从视频关键帧中提取的局部特征采用一种新的自适应局部敏感哈希算法进行索引.通过样本学习并设定一些参数,使检索过程不需要进行高维距离计算,从而有效的提升了处理速度.检索得到的特征向量备选集经过特征过滤和两层匹配方法完成重复视频的检测流程.在标准数据集上的实验表明,与其他最新的方法相比,本文提出的算法有效的提高了大规模重复视频的检测速度.With the increasing number of multimedia applications in commercial markets, large scale near-duplicate video detection is very desirable for web video processing, especially the computational efficiency is essential for practical applications. In this paper, we present a computationally efficient algorithm based on multi-layer video content analysis. Local features are extracted from key frames of videos and indexed by a novel adaptive locality sensitive hashing scheme. By learning several parameters, fast retrieval in the new hashing structure is performed without high dimensional distance computations and achieves better real-time retrieving per- formance compared with other state-of-the-art approaches. Then a descriptor filtering method and a two-level matching scheme is per- formed to generate a relevance score for detection. Experiments on near-duplicate video detection tasks including various transformed videos demonstrate the efficiency gains of the proposed algorithm.

关 键 词：重复视频检索 局部敏感哈希 SURF 多媒体内容分析 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]