Optical flow based model for scene cut detection

Cut detection is a fundamental operation for video sequence segmentation. This paper presents an optical flow based model with linear prediction to realize cut detection in various transition conditions. Each frame of a video sequence is divided into 4×4 non-overlapping blocks. This model initiates the cut detection by backward block searching for a match. Three continuous frames with the same matched block and the fourth predicted frame are grouped as a subsequence. The linearly predicted non-overlapping block locations in the fourth frame are defined as a cut detectable region. The forward search for a match is performed between the third and fourth frames in the cut detectable region. Mismatched blocks indicate a scene change in these block locations, namely, a cut. This algorithm has a better performance for detecting cuts in quick scene changes, static scenes, and scenes with slow motion compared to techniques reported in the literature