基于时空组合特征的视频拷贝快速检测

为了有效提升大规模数据集下视频拷贝检测的速度,提出了一种基于时空组合特征的视频拷贝快速检测算法。首先抽取视频的时间序列特征和空间分布特征,然后采用粗略匹配和精确匹配相结合的二级匹配框架进行拷贝检测,并在每一级匹配过程中都加入了"尽早停止"策略,以便尽可能快地过滤出非拷贝视频,从而降低在大规模数据集下的计算复杂度。实验结果表明,与已有的算法相比,所提出的快速检测算法取得了较高的检测精度,并大幅提升了检测速度。

Rapid Video Copy Detection Based on Spatial-Temporal Feature

In order to improve the detecting efficiency of video copy under large scale video data,a rapid video copy detection approach is proposed based on a novel spatial-temporal feature.First,a combined feature set,including time-sequence feature and spatial-distribution feature,is extracted to describe the video data.Second,a two-step(coarse and fine) similarity evaluation method with early-stopping scheme is proposed to detect the video copies.The coarse evaluation scheme and early-stopping method filter many n...