Real-time entropic unsupervised violent scenes detection in Hollywood movies - DYNI @ MediaEval Affect Task 2011

State of the art of the high level feature detectors, as violent scene detectors, are supervised systems. The aim of our proposition is to show that simple non supervised confidence function derived from straightforward features can perform well compared to nowdays supervised systems for this kind of hard task. Then, we develop a violent event detector independent of the kind of movies based on our previous research on basic efficient entropic movie features. We propose an entropic audiovisual confidence computed as the average of the entropies of some simple visual and acoustic features. In a first approach, we develop our system for uniform false alarm and missing costs, which is not optimal according to the official campaign criterion. However, the usual Fmeasure metrics indicates that our system is the second best among the five other -supervisedsubmitted systems.