Multi-modal Multiple-Instance Learning And Attribute Discovery With The Application To The Web Violent Video Detection

Due to the openness of the Internet, violent videos can be spread widely and easy to access. These videos are harmful to society and have caused serious consequences, which are often reported on line. Therefore, there is absolutely a need to propose a method to detect violent videos and guarantee the safety of the information on the Internet. Therefore, in this paper, we focus on the violent video classification and the fusion of the text, image and audio,as well as its application to violent video filtering. The main contributions of are summarized as follows:1. Extract the expression attribute vocabulary through video introduction and user review,and then establish a diagram that represents the relationship between attributes, this relationship behind the launch of audio and image features.2. Extract image and audio features from video, and then extract text features from video introduction and user review.3. Detect violent video by using Multi-instance learning algorithm, they have some commons. If a instance is positive, the bag will be positive, which is similar to violent video.If a video camera includes violent content, the video is violent.4. Design an efficient instance selection technique by utilizing text information to speed up the training process without compromising the performance. Experiments show that the approach not only consider the practicality, speed, and taking into account the performance of the classifier and the classification of violent video for the special requirements of this particular area.