PKU-IDM @TRECVID2011 CBCD: Content-Based Copy Detection with Cascade of Multimodal Features and Temporal Pyramid Matching.

Menglin Jiang,Shu Fang,Yonghong Tian,Tiejun Huang,Wen Gao
2011-01-01
Abstract:Content-based copy detection (CBCD) is drawing increasing attention from both academia and industry as an alternative technology to watermarking for video identification and copyright protection. In this paper, we present a comprehensive method for detecting copies subjected to complicated transformations in a large video corpus. Basically, two core techniques are employed by our method. One is multimodal feature representation organized in a cascade architecture, which exploits the complementary characteristics of audio features, global and local visual features to keep robust to a wide range of transformations and meanwhile preserves efficiency as far as possible. The other is Temporal Pyramid Matching (TPM), which fuses frame-level similarity search results into sequence-level matching results. We have submitted two runs, i.e. “PKU-IDM.m.balanced.cascade” & “PKU-IDM.m.nofa.cascade”. Official results demonstrate that the proposed approach achieved excellent NDCR and competitive Mean F1 at the cost of median Processing Time.
What problem does this paper attempt to address?