The detection of significant scene changes in a video sequence is discussed in this paper, with a focus on those in a particular cut (abrupt scene change), fade out or in (gradual transition to or from a constant monochromatic image), and flash (luminance surge in a single video frame). To perform this detection, the video is first transformed into data called visual rhythm, thanks to a reduction of each 2D video frame into 1D data. This 2D to 1D reduction can be performed either by a sampling (for example, along a diagonal), or by taking the histogram of the frame. In both types of visual rhythms (by sampling and by histogram), scene changes appear as specific features, which can be extracted by morphological and topological operators. The authors have also included a statistical analysis of the rates of false positives and false negatives of their method, as compared to previous approaches.
This visual rhythm idea of simplifying the sequence into a one comes from previous papers referenced by the authors. The originality of the present work lies in the use of morphological and topological operations in order to analyze the changes in visual rhythm.