NII-UIT at MediaEval 2015 Affective Impact of Movies Task

Abstract

Aective Impact of Movies task aims to detect violent videos and aective impact on viewers of that videos [9]. This is a challenging task not only because of the diversity of video content but also due to the subjectiveness of human emotion. In this paper, we present a unied framework that can be applied to both subtasks: (i) induce aect detection, and (ii) violence detection. This framework is based on our previous year’s Violent Scene Detection (VSD) framework. We extended it to support aect detection by training dierent valence/arousal classes independently and combine them to make the nal decision. Besides using internal features from three dierent modalities: audio, image, and motion, in this year, we also incorporate deep learning features into our framework. Experimental results show that our unied framework can detect violent videos and its aective impact with a reasonable accuracy. Moreover, using deep features can signicantly improve the detection performance of both subtasks.