Video Affective Content Recognition Based on Film Grammars and Fuzzy Evaluation

Affective content analysis is an unsolved technical problem in sophisticated video retrieval and high-level applications. In order to recognize emotion types of video scenes, a new algorithm, which composes of two sub-models, is proposed. Firstly, the sub-models of two low-level features extraction are built up based on the film grammars. Secondly, a classification sub-model of scene emotion is presented. This sub-model includes two functions: fuzzy relation matrix computed by fuzzy membership functions which are built from a large number of fuzzy experiments, and affective type decision function based on the maximizing decision making of fuzzy evaluation. Experimental results show the proposed algorithm is feasible and achieves a high recognition accuracy which exceeds 80 percent.