1 – Shot Boundary Detection Our shot boundary work was based upon three core techniques. The first, histogram similarity, involved construction of color histograms for each frame, where the color space was compressed to RGB values of three bits each, yielding a 512 bin histogram and also represented in HSB space. We have also included an averaging filter that is very useful in reducing the effects of color ‘jitter’ in noisy video. The second technique (distance) involves computing for a pair of frames the aggregate color distance for pixel pairs (having the same location in their respective frames) and then normalizing this value by the dimensionality of the frame. This technique has proven quite useful in avoiding false positives due to dramatic, but localized, color shifts between frames. The third technique (edge) involves edge generation from each frame (using the ImageJ library) and applying the distance metric to pairs of now-monochrome frames. This yields a measure of gross movement occurring between frames that is tunable by how aggressively we erode edges before calculating distances between frames. Our composite HSB technique first does a histogram-based cut detection and then overlays that with an averaged HSB gradual detection, with graduals trumping any contained cuts.
[1]
Jean-Luc Gauvain,et al.
The LIMSI Broadcast News transcription system
,
2002,
Speech Commun..
[2]
Jonathan G. Fiscus,et al.
A post-processing system to yield reduced word error rates: Recognizer Output Voting Error Reduction (ROVER)
,
1997,
1997 IEEE Workshop on Automatic Speech Recognition and Understanding Proceedings.
[3]
Ramin Zabih,et al.
A feature-based algorithm for detecting and classifying scene breaks
,
1995,
MULTIMEDIA '95.
[4]
Dong-Jun Park,et al.
Experiments in Boundary Recognition at the University of Iowa
,
2003,
TRECVID.
[5]
Thomas G. Dietterich.
An Experimental Comparison of Three Methods for Constructing Ensembles of Decision Trees: Bagging, Boosting, and Randomization
,
2000,
Machine Learning.