The researches on horror network information filtering not only promote the development of the network content security and human emotion research in the field of cognitive, but also play an important role in building a harmonious environment and maintaining social stability. In this paper, we focus on the study of horror videos. As far as we know, there is nearly no research on the identification of horror videos before us, only image and video emotion analysis research is associated with the identification of horror video. To solve the problem of the horror video recognition, we should clear up two points as follows: (1) How to determine the salient emotion area in the image and extract the visual features with emotion? (2) How to effectively integrate audio and video emotional information to identify the horror video? We carry out series of research work from the feature level to the framework level, the main work is as follows: 1. Make a survey about video structure algorithm, and implement the method for shot boundary detection which relies on the mutual information (MI) and the joint entropy (JE) between the frames. 2. Search and research a large number of literature on image emotion analysis and audio emotion analysis, extract the relevant audio and video emotional features,furthermore introduce the results of emotion analysis into the horror video recognition. 3. Introduce the color emotion and color harmony theory based on psychological experiments and propose a new color emotional feature with the high-level semantics.Experiments show that the proposed color emotion features can improve the recognition result of horror videos. 4. Propose the framework for horror video identification based on emotional perception and achieve promising experimental results. 5. Make the structural analysis for video, then multi-instance learning is applied to horror video recognition in innovative way. Experimental results show that the multi-instance learning method is well suited to address the problem of horror video recognition. 6. Build an experimental database. In a word, in this thesis, we have made a lot of fruitful attempts and significant progresses on horror movie scene recognition.
修改评论