空洞卷积网络下微表情实时识别方法仿真

Simulation of Real-Time Recognition Method of Microexpression under Empty Convolution Network

微表情与普通面部表情不同,是一种面部动作变化微弱、持续时间极短的面部活动,且需要在视频中分析,因此特征提取较为困难。为了解决上述问题,提出基于空洞卷积的实时微表情识别算法。通过混沌蛙跳算法,对人脸微表情图像增强处理。采用时间差值法和局部二值法,提取人脸特征信息。结合空洞卷积构建卷积神经网络,并将提取到的人脸特征输入到构建的卷积神经网络,完成人脸微表情的实时识别。实验结果表明,所提方法的实时微表情识别准确率在97%以上,且识别时间短,说明所提方法具有较好的实际应用价值。

Micro-expression is a kind of facial activity with weak changes and extremely short duration,which needs to be analyzed in the video,so it is difficult to extract these features.Therefore,a real-time micro-expression recognition algorithm based on dilated convolution was proposed.At first,the chaotic leapfrog algorithm was adopted to enhance the facial micro-expression image.Then,the time difference method and local binary method were used to extract facial feature information.Moreover,a convolution neural network was constructed by dilated convolution.Meanwhile,the extracted face features were input into the convolution neural network,thus completing the real-time recognition for facial micro-expressions.Experimental results show that the proposed method has more than 97%accuracy in real-time micro-expression recognition and less recognition time,indicating that the proposed method has good practical application.