Dialogue Scenes Detection in MPEG Movies: A Multi-expert Approach

In this paper we propose a method for the detection of dialogue scenes within movies. This task is of particular interest given the special semantic role played by dialogue based scenes in the most part of movies. The proposed approach firstly operates the segmentation of the video footage in shots, then each shot is classified as dialogue or not-dialogue by a Multi-Expert System (MES) and, finally, the individuated sequences of dialogue shots are aggregated in dialogue scenes by means of a suitable algorithm. The MES integrates three experts which consider different and complementary aspects of the same decision problem, so that the combination of the single decisions provides a performance that is better than that of any single expert. While the general approach of multiple experts is not new, its application to this specific problem is interesting and novel and the obtained results are encouraging.