Video shot-boundary detection using singular-value decomposition and statistical tests

We deal with video shot-cut detection in digital videos using the singular-value decomposition (SVD). SVD is performed on a matrix whose columns are the 3D frame color histograms. We have used SVD for its capabilities to derive a refined low-dimensional feature space from the high-dimensional raw feature space, where similar video patterns are placed together and can be easily clustered. After SVD is performed, a two-phase process is employed to detect the shots. In the first phase, a dynamic clustering method is used to create the frame clusters. In the second phase, every two consecutive clusters, obtained by the clustering procedure, are tested for a possible merging in order to reduce false shot-cut detections. In the merging phase, statistical hypothesis testing is used. The detection technique was applied to several TRECVID video test sets that exhibit different types of shots and contain significant object and camera motion inside the shots. We demonstrate that the method detects cuts and gradual transitions, such as dissolves and fades, with high accuracy. (c) 2007 SPIE and IS&T.