An adaptive video shot segmentation scheme based on dual-detection model

Efficient segmentation of the video shots is the important and foundational work for the research of video content retrieval and analysis. A video shot segmentation scheme based on a dual-detection model is proposed, which includes the pre-detection and re-detection processes. The concepts of uneven blocked color histogram difference and uneven blocked pixel value difference based on human visual features are introduced, which are used as the main descriptors of the pre-detection process to enlarge the importance of central areas and to reduce the noises of background movements and logos. And the adaptive binary search method is introduced to fast detect boundaries with a time complexity of O(log2n). In the re-detection round, the scale invariant feature transform is applied to re-detect boundaries so as to improve the detection precision rate. Experiments show that this algorithm can improve both the recall rate and the precision rate of video shot boundary detection, especially for gradual shot boundaries.