Fast Video Copy Detection

With digital video content production and distribution continuing to grow, content-based copy detection (CBCD) has been actively studied for a wide range of applications that include searching, multimedia linking, and protecting copyrighted content. Based on content alone, CBCD attempts to identify segments in a query video that are copies from a reference video database. A copy is not an exact duplicate but, in general, either a transformed or a modified version of the original document that remains recognizable. This task is challenging since two copies might be visiaully dissimilar, as shown in the figure below.

We propose an edit-distance-based approach that has the potential for large-scale CBCD applications. We first formulate a local alignment problem between two sequences and extend previous edit-distance-based approaches to compare video segments of all possible lengths. The main contribution of this work is the highly efficient matching process between a query video and a video database which is achieved by a fast local alignment method along with a dedicated index structure that provides detection acceleration at both the clip and frame levels. We demonstrate the effectiveness and efficiency of this method using the MUSCLE VCD benchmark.

Mei-Chen Yeh and Kwang-Ting Cheng. Fast Visual Retrieval Using Accelerated Sequence Matching, to appear in IEEE Transactions on Multimedia.

Mei-Chen Yeh and Kwang-Ting Cheng. Video Copy Detection by Fast Sequence Matching. ACM International Conference on Image and Video Retrieval 2009 (ACM CIVR 09), July 8-10, Island of Santorini, Greece.

<back to research>