Audio-Visual Content-Based Violent Scene Characterisation

Abstract

In this thesis, I show a novel strategy to portray and list violent events in TV dramas and movies. Our objective is to distinguish violent scenes from the normal scenes and confine brutal occasions inside a film to bolster "abnormal state" feature indexing. Specially, I have investigated multiple “audiovisual” signals to create a perceptual connection for reasonably important vicious scene recognizable proof. Potential applications are programmed obstructing of violent scenes in motion pictures observed by kids, hiding viciousness utilizing information filtering or data sifting and genre classification of advanced feature database.