中文摘要：

本文中特定敏感视频是指恐怖和暴力视频,现有的特定敏感视频识别算法或是忽略了视频的多种上下文结构信息;或是忽略了各种特征间潜在的依赖关系.因此,本文提出了一种基于多种上下文结构与线性融合的特定敏感视频识别方法,首先针对某种视频提取多种有效特征,并获取镜头间的上下文结构信息;然后,在每一个特征空间中利用上下文结构训练一个SVM分类器;最后,获取不同特征间的依赖关系,采用线性依赖模型融合多个分类器的结果,提高视频的识别率.在特定敏感视频库上的实验结果验证了该方法比现有的其它算法有更好的性能和稳定性.

英文摘要：

Along with the ever-growing Web, specified sensitive videos including horror videos and violent videos are disseminated over Intcrnet and have threatened children＇ s psychological health. It is necessary to effectively recognize and filter out these videos. So far, the existing recognition methods for specified sensitive videos either ignore the multiple contextual information among the shots, or ignore the dependent relationship among the multiple features. This paper presents a novel recognition method for specified sensitive video based on multi-context construction and linear fusion. First, multiple effective features are extracted and the multiple contextual structure graphs are constructed for the shots in one video clip. Then, a SVM classifier is trained using contextual information in each feature space. Finally, a linear dependency model is learnt to fuse multiple classifiers obtained in different feature spaces. Experiments on a specified sensitive video dataset demonstrate that the performance of our method is superior to the other existing algorithms.