(ppt) IBM Content Based Copy Detection System for TRECVID 2009.pdf

? 2009 IBM Corporation IBM Content Based Copy Detection System for TRECVID 2009 Speaker: Matt Hill On behalf of: Jane Chang, Michele Merler, Paul Natsev, John R. Smith ? 2009 IBM CorporationIBM Research
We explored 4 complementary approaches for video fingerprinting: – Two frame-based visual fingerprints (color correlogram and SIFTogram) – Two temporal sequence-based fingerprints (audio motion activity)
Key question: How far can we go with coarse-grain fingerprints? – Focus on common real-world transforms typical for video piracy detection – Focus on speed, space efficiency, lack of false alarms System Overview Videosi S e g m e n t - b a s e d F i n g e r p r i n t s S e g m e n t - b a s e d F i n g e r p r i n t s Motion activityti ti it Audio activityi ti it SIFTogramI r Color correlograml r rr l r F r a m e - b a s e d F i n g e r p r i n t s F r a m e - b a s e d F i n g e r p r i n t s Fingerprints Fusion Mean-Normalized Video-Only Run - r liz i - l Median-Normalized Video-Only Run i - r liz i - l Mean-Normalized Audio-Video run - r liz i - i r Median-Normalized Audio-Video Run i - r liz i - i ? 2009 IBM CorporationIBM Research We focused on CBCD transforms that represent typical video piracy scenarios (i.e., ignore PIP and post-production edits)
T2: Picture in picture Type 1 (The original video is inserted in front)
T3: Insertions of pattern
T4: Strong re-encoding
T5: Change of gamma
T6: Decrease in quality -- This includes choosing randomly 3 transformations from the following: Blur, change of gamma, frame dropping, contrast, compression, ratio, white noise
T8: Post production -- This includes choosing randomly 3 transformations from the following: Crop, Shift, Contrast, caption (text insertion), flip (mirroring), Insertion of pattern, Picture in Picture type 2 (the original video is in the background)
T10: change to randomly choose 1 transformation from each of the 3 main categories. We focused on the