基于人脸多模态的视频分类算法的设计与实现

Design and implementation of video classification based on face multimodality

文中提出了利用视频内部人物情感的数据实现对视频智能化分类的功能。给出了人脸检测、人脸矫正、人脸多模态识别算法,视频关键场景确定算法,人物情感数据分析算法,并使用opencv和caffe实现算法并验证其可行性。基于以上的算法处理后,可以完成自动对短视频的情感分类,分类结果如积极、消极还是中间过渡情感等。其中表情识别准确率高于baseline算法4.5个百分点,同时对视频分类算法准确率基本居于60%以上。

This paper proposes the function of video intelligent classification based on the data of emotion inside the video.This paper presents face detection,face correction,facial multimodal recognition algorithm,video key scene determination algorithm,the human emotion data analysis algorithm,and use opencv and caffe to achieve the algorithm and verify its feasibility.After the above processing,we can complete the emotional classification of short videos automatically,the classification results such as positive,negative or intermediate transitional emotions.Among them,the emotion recognition accuracy rate is 4.5%higher than the baseline algorithm,and the accuracy of the video classification algorithm is basically above 60%.