Abstract
The news very often reports on the bad influence that violence in movies has on children and young adults. Frequently, censors are employed to identify and remove such content. This is a time consuming and expensive task and therefore an automatic process would be beneficial.
Software which solves this task in a fully automated fashion is not imaginable at this point in time because there are many different forms of violence and people generally disagree on whether a film scene is violent or not. Still, in this work a method is proposed to classify film scenes as either containing or not containing violent content in a computer assisted manner. For this task it is necessary to extract features that are characteristic for violent scenes. Then a classifier labels film scenes in violent and nonviolent scenes with respect to the chosen features.
The output of the algorithm are all scenes that contain violence. The research prototype was implemented in the Matlab environment. To realize this prototype several experiments with different retrieval methods, various features and multiple classifiers were made. The tested methods and features that gave the best results were incorporated in the prototype. First results of the prototype evaluation are encouraging.
Software which solves this task in a fully automated fashion is not imaginable at this point in time because there are many different forms of violence and people generally disagree on whether a film scene is violent or not. Still, in this work a method is proposed to classify film scenes as either containing or not containing violent content in a computer assisted manner. For this task it is necessary to extract features that are characteristic for violent scenes. Then a classifier labels film scenes in violent and nonviolent scenes with respect to the chosen features.
The output of the algorithm are all scenes that contain violence. The research prototype was implemented in the Matlab environment. To realize this prototype several experiments with different retrieval methods, various features and multiple classifiers were made. The tested methods and features that gave the best results were incorporated in the prototype. First results of the prototype evaluation are encouraging.
Reference
Hörhan, M. (2010). Automatische Erkennung von Gewaltszenen in Filmen [Diploma Thesis, Technische Universität Wien]. reposiTUm. http://hdl.handle.net/20.500.12708/161373