Skip to Main Content
Acoustic features are robust and powerful in video description, but not fully exploited for the emerging Content-Based video Copy Detection (CBCD) methods. To solve this discrepancy, this paper proposes a new CBCD approach using audio spectral features compared to existing visual content based methods. The proposed method incorporates three stages: (1) Extraction of spectral descriptors including centroid and energy; (2) Integration of resultant features to compute highly informative spectral descriptive words; (3) Utilization of clustering approach to speed up the similarity matching process. The results tested on TRECVID-2008 dataset, demonstrate the improved detection accuracy of proposed method (up to 27.845%) compared to reference methods against various transformations such as fast forward, slow motion, mp3 compression, and multiband companding.