基于大脑响应特征的视频情感分类算法

Video Emotion Classification Algorithm Based on Brain Response

视频情感识别是计算机视觉的研究热点,由于认识到人类本身才是情感产生的源头,近来,利用人类自身的大脑响应等生理特征对视频所包含的情感进行识别,即隐性情感识别成为研究重点。然而,目前利用脑电图信号对音乐视频愉悦度的识别率仍不能令人满意,原因在于未能从大量的脑电图数据中获取到有效的分类特征。为了进一步提高识别准确率,在DEAP数据库中,不采用传统的脑电图时域和频域特征,而是利用数据标准化以及特征选择方法从脑电图时间序列信号中直接提取有效特征,从而提取到脑电图信号中具有较高分类能力的特征,并将得到的脑电图特征用于音乐视频分类实验中,结果表明,相对于传统方法,可以大大提高脑电图信号对音乐视频愉悦度识别率。

Video emotion recognition is the research focus of computer vision. Recently,implicit emotion tagging which uses human physiological such as brain signal from human to recognize video emotion becomes research hotspots,due to that researcher realized the human is the source of the emotion. However,the existing accuracy rate for valence recognition of music video is still far away from satisfactory,due to the effect feature has not been extracted from many EEG data. In order to improve the precision,based on the DEAP dataset,instead of using traditional time and frequency domain feature of the EEG,this paper uses the data standardization and feature selection method to select the useful features from EEG time frequency directly. Finally,the feature with higher classification ability is extracted from EEG data. The selected EEG features then are used to classify the music video.Experimental result demonstrated that our algorithm performed much better than the traditional algorithm in the valence classification of music video.