一种基于SimHash的海量视频检索方法  

A Massive Video Retrieval Method based on SimHash

作　　者：楚敏南 罗新高[1] 白煜华 

机构地区：[1]湘潭大学信息工程学院智能计算与信息处理教育部重点实验室 [2]75140部队

出　　处：《科技与创新》2015年第18期9-11,共3页Science and Technology & Innovation

摘　　要：针对海量视频检索,提出了一种基于SimHash的视频相似性检索方法。该方法的视频特征提取部分首先采用视觉词袋模型将视频关键帧表示为词袋模型向量,然后对高维词袋模型向量建立鲁棒的压缩二值SimHash签名;视频相似帧查找部分首先置换SimHash签名库,并排序得到多张签名表,然后在多张签名表中按照数据量合理利用BloomFilter算法精确匹配签名表的置换部分,进而根据精确匹配的结果高效查找汉明距离小于阈值的签名,最后利用查找到的签名对相关视频进行相似度计算,排序得到相似视频的查询结果。针对CC_WEB_VIDEO公开数据集的实验表明,该方法对大规模视频的快速检索是非常有效的。In this paper, a video similarity retrieval method based on SimHash is proposed in this paper. The method of video feature extraction part, the bag of visual words model video key frame representation for the bag of words model vector, then the high-dimensional bag of words model vector to establish a robust compression of binary SimHash signature; video similarity search of a flame first replacement SimHash signature repository, side by side in order get a signature table, then in a signature table according to the amount of data, reasonable use bloom f^lter algorithm exactly match the signature of the replacement part and according to the exact match resultsl find the Hamming distance is less than the threshold signature, finally find the signature of the video similarity calculation, sort query results of similar video. Experiments with the CC WEB_VIDEO open data set show that the method is very effective for the rapid retrieval of large-scale video.

关 键 词：视频检索 视觉词袋 SimHash BLOOMFILTER 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]