A Robust Instance-based Video Search Method by Re-ranking for Multiple Visual Codebooks

With the development of Internet and computer technology, the digital videos increase explosively, so how to get the interesting video clips from the massive video dataset quickly and efficiently has become an urgent problem in the field of information retrieval (IR). In this paper, a new Instance-based video search (INS) engine is proposed to solve the urgent problem. Firstly, video key-frames are extracted and grouped from queries and video dataset. Secondly, robust low-level visual features are extracted and projected to multiple visual codebooks. Finally, after similarity computing and feature fusion, a re-ranking scheme is implemented to vote the final search results. The proposed framework was evaluated at TRECVID 2011 on instance search task(INS), and achieved the 2^nd place among 47 participants around the world, which indicated the effectiveness of our system.