基于时空灰度序特征的视频片段定位算法

Video Clip Identification Algorithm Based on Spatio-Temporal Ordinal Measures

基于灰度序特征的视频片段定位算法是解决视频片段定位问题的典型算法.这类算法存在的不足是:特征的唯一性表示能力不够,使得在召回率较高的情况下,定位检索的精度下降得较快;二次多项式级的时间复杂度使得响应时间过长,并对查询视频长度敏感.针对上述两个问题,提出了一种基于时空灰度序特征的视频片段定位算法,其关键步骤包括:(1)在精确定位之前,通过引入线性时间复杂度的基于时空二值模式直方图特征(spatio-temporal binary pattern histogram,简称STBPH)的实时过滤算法以及基于二值时间灰度序特征(binary temporal ordinal measure,简称BTOM)的快速过滤算法,大幅度减少精确定位阶段需要进行比较的候选视频片段个数;(2)在精确定位阶段,通过引入唯一性表示能力更好且保持了较好鲁棒性的时空统一灰度序特征(joint spatio-temporal ordinal measure,简称JSTOM)进行序列匹配,显著提高了定位检索的精度.实验结果表明,该算法能够快速、准确地进行视频片段定位,大幅降低了对查询视频长度的敏感度.

Many state-of-the-art video clip identification algorithms are based on ordinal measures. However, they still have two problems： The weak uniqueness of video signature makes the precision decreases quickly as recall increases high enough; Quadratic-time complexity makes the response time too long and sensitive to the length of query video. To address these two problems, this paper proposes a video clip identification algorithm based on spatiao-temproal ordinal measures. The key steps are： （1） Before the accurate identification starts, it employs a linear-time complexity real-time filtration method based on spatio-temporal binary pattern histogram （STBPH） and a fast filtration method based on binary temporal ordinal measure （BTOM） to filter out most candidate video clips in target video; （2） During the accurate identification process, it utilizes joint spatio-temporal ordinal measure （JSTOM） which is more unique and robust in improving the precision. Experimental results show that the approach improves the precision significantly and is very efficient and insensitive to the length of query video.