Video Comparison Using Tree Matching Algorithms

This paper proposes a video structure matching framework based on a tree matching algorithm which can be used as an effective video retrieval tool in a media management system. We define two types of specialized tree matching algorithms. The first is the non-ordered tree matching and the second is the ordered tree matching. The first algorithm is not constrained by the temporal ordering of the video, whereas the second algorithm takes the temporal ordering into account. Moreover, these two algorithms are special since they both match a well-structured tree-like video model having the same tree depth. Our experiments on a small set of various videos demonstrate that the proposed tree matching algorithms produce similar ranking results to what human will produce.