An effective dissolve detector using spatio-temporal slice

The detection of edits in a video sequence is the first step in video analysis, which segments a video into its basic components. Spatio-temporal slice analysis is an effective method for video partitioning because it can detect and classify different scene breaks. In a spatio-temporal slice, cut and wipe can be detected successfully based on measuring the changes of the color-texture properties of the slices. Dissolve can be measured by means of the parabolic variance curve (PVC) method. However, the statistical information extracted from the horizontal, vertical, and diagonal slices is not enough to show the PVC features. Thus, Support Vector Machine (SVM)-based dissolve detector was proposed, which extracts features based on the Gabor wavelets from a spatio-temporal slice and then identifies dissolves by means of the SVM-based classifier. However, this method is computationally intensive. In our method, we propose an efficient dissolve detector based on the spatio-temporal slices by using three simple second-order filters. Based on the linear estimation of the successive frames in a video shot, dissolve and static scenes exhibit different patterns in the temporal dimension. By applying the three simple filters, we can identify dissolves with arbitrary lengths accurately. Experiments based on the MPEG-7 standard sequences show encouraging results.