Violence in movies and its relation to children’s behavior has been a subject of many social studies, specially in today’s world where parents have less control on the multimedia contents accessed by their children. This paper presents a study on the automatic detection of violence on movie excerpts using machine-learning algorithms. The proposed approach explores audio and visual features extracted from the movie excerpts combined with the widely used machine learning classifiers – Support Vector Machine (SVM) and Neural Network – in order to detect the presence of violence in a movie excerpt. The impact of using audio and visual features, independently or combined, is analyzed. After experimenting different combinations of audio and video features, interesting conclusions have been drawn: the use of audio features seems to be imperative in this task and, from the video signal, the most representative features seem to be the ones related to motion.

--