Key frame vector and its application to shot retrieval

Abstract

This paper proposes a video representation method named key frame vector (KFV) to support video shot retrieval. By considering temporal correlation between frames reflects the relations between visual contents of frames, and video content weighting has influence on the video comparison performance, the proposed KFV incorporates the temporal correlation and spatial relations between frames, as well as content weighting. It employs a block-based frame comparison scheme, and uses EMD to compare shots represented by KFVs. The proposed algorithm has been demonstrated to be effective in comparison with state-of-the-arts methods. Copyright 2009 ACM.