NTT Communication Science Laboratories at TRECVID 2010 Content-Based Copy Detection
Ryo Mukai, Takayuki Kurozumi, Kaoru Hiramatsu, Takahito Kawanishi, Hidehisa Nagano, Kunio Kashino · 2010
In this paper, we describe our approaches that were tested in the TRECVID 2010 Content-Based Copy Detection (CBCD) task. We introduce a method consisting of a feature degeneration and sparse feature selection process for video detection tasks, which is highly robust as regards video signal distortion. For audio detection tasks, we adopt a method based on spectral partitioning to cope with additive interfering sounds. Both methods are key techniques for our Robust Media Search (RMS) technology, which has already been deployed for various commercial services. Evaluation results show the effectiveness of our methods. 1