利用等价关系理论进行视频片段检索的方法

An Efficient Method for Video Clip Retrieval Using Equivalence Relation Theory

视频片段检索是基于内容的视频检索的主要方式,可是现有的片段检索方法大多只是对预先分割好的片段进行检索。为了从连续的视频节目中自动分割出多个相似的片段,提出了一种新的有效的视频片段检索方法,并首次尝试将等价关系理论应用于视频片段的检索.该方法首先用等价关系理论定义了片段匹配函数,同时采用滑动镜头窗自动分割出多个真正相似的片段;然后把等价类映射为矩阵表达形式,再通过矩阵的特性来度量影响片段相似度的不同因子,实现了相似片段的排序。实验结果表明,该方法能够一次性快速准确地从连续视频库中自动分割出与查询片段相似的多个片段。

Video clip retrieval plays a critical role in the content-based video retrieval. However, existing clip retrieval methods mostly focused on retrieving similar video clips from pre-segmented clips. To automatically segment and retrieve similar video clips from a continuous video database, this paper presents an efficient method for video clip retrieval in which equivalence relation theory is applied to video clip retrieval for the first time. Matching function is defined in terms of equivalence relations corresponding with shots of given video clip. Several similar video clips are automatically segmented from video database using sliding shot window according to matching degree between two video clips. Afterwards, equivalence classes are mapped into matrix representation. Therefore various factors are computed to rank the similarity of the selected video clips by characters of the matrix. Experimental results showed that the method could segment similar video clips from continuous video database quickly, exactly and automatically.