Fast video search using scene durations

This paper presents a fast video search method that uses the durations of scenes of a query clip to find similar clips in an existing large database of videos. Using scene durations is a very fast method; however it returns a large number of false candidates and will not find the correct candidates if some scene changes are missed in the query clip or the database videos. The first problem can be solved by further refining the results of the scene duration search by comparing these results with the query clip using other, more detailed characteristics like brightness histogram. The second problem can be solved by introducing flexibility in the duration and existence of a scene. Numerical results show the efficiency of the scene duration search and its robustness against drawbacks of the basic scene detection algorithm. The presented methodology is used in the video search part of project DIVAS of the EU Framework Programme.