Scene change detection for uncompressed video

In this paper, we present a scene change or shot boundary detection method based in the changes in entropy of differences between uncompressed video frames. As in the uncompressed domain, cues for scene or shot boundaries are not available, detecting video content features is a non-trivial and typically requires additional complexity in the evaluation. The entropy presents a metric for the complexity of information. Used on the differences between video frames, the entropy is able to measure the complexity of changes. We find that due to content dependency, however, the relative entropy changes in the sequence of video frames is a better indicator for detection. An evaluation of the presented approach finds that detection for a combination of video test sequences can be reliably performed using the U component of uncompressed YUV 4:2:2 video only.