基于情感特征融合的视频恐怖场景检测算法

Detection algorithm for video horror scene based on affective attention feature fusion

为有效识别网络视频中的恐怖场景,避免未成年人在互联网上访问不良信息,提出了一种自上而下的融合方法,有效地融合了恐怖场景识别中的音频、视觉和情感特征.首先由视频场景的不同特征构成基于情感词汇的词包模型;然后选择部分特征作为显著性特征,为其余特征赋予权重;最后,依据自上而下的特征融合策略所生成的特征向量进行恐怖场景识别.试验结果表明,该方法能有效识别恐怖视频.

In order to effectively identify the horror scene in the network video and avoid the underage people from accessing the objectionable contents in the social media.This paper proposes a top-down fusion method to effectively fuse the audio,visual and emotional features for horror scene recognition.First,different features of the video scene can be viewed as a bag of affective words.Then,one kind of feature is selected as saliency cue to give weights on the features left.Finally,the feature vector generated by the top-down fusing strategy is used for horror scene recognition.According to the experimental results,the proposed method can effectively deal with the horror video scene recognition.