摘要
传统人脸表情识别主要基于人工提取特征,其存在算法鲁棒性较差、易受人脸身份信息干扰等问题,以及传统卷积神经网络不能充分提取人脸表情特征的现状。对此提出一种基于多特征融合密集残差卷积神经网络的人脸表情识别。该方法能够充分利用神经网络中每层的特征,在密集块中,对于每一个卷积层,其前面所有卷积层的输出都将作为本卷积层的输入。然后将每个密集块的输出送入到全连接层中进行特征融合,经过Softmax分类器分类。在CK+和FER2013数据集上进行多次实验,与传统的机器学习方法相比,该方法具有较高的准确率与较强的鲁棒性。
Because the traditional facial expression recognition is mainly based on the artificial extraction of features, the robustness of the algorithm is poor, and it is easy to be interfered by the face identity information. Traditional CNN cannot adequately extract facial expression features. We proposed a facial expression recognition based on multi-feature fusion dense residual CNN. This method could make full use of the characteristics of each layer in the neural network. For each convolution layer in the dense block, the output of all convolution layers in front of it would be the input of this convolution layer. Then the output of each dense block was fed into the full connection layer for feature fusion, and classified by Softmax classifier. Experiments were performed on CK+ and FER2013 data sets. Compared with traditional machine learning, our method has higher accuracy and stronger robustness.
引文
[1] Mollahosseini A,Chan D,Mahoor M H.Going deeper in facial expression recognition using deep neural networks[C]//2016 IEEE Winter Conference on Applications of Computer Vision(WACV).Lake Placid,NY,USA:IEEE,2016:1-10.
[2] Xie S,Hu H.Facial expression recognition with FRR-CNN[J].Electronics Letters,2017,53(4):235-237.
[3] Pramerdorfer C,Kampel M.Facial expression recognition using convolutional neural networks:state of the art[EB].arXiv preprint arXiv,1612.02903,2016.
[4] Lu P,Li B,Shama S,et al.Regularizing the loss layer of CNNs for facial expression recognition using crowd sourced labels[C]//2017 21st Asia Pacific Symposium on Intelligent and Evolutionary Systems(IES).Hanoi,Vietnam:IEEE,2017:31-36.
[5] 李勇,林小竹,蒋梦莹.基于跨连接 LeNet-5网络的面部表情识别[J].自动化学报,2018,44(1):176-182.
[6] 何志超,赵龙章,陈闯.用于人脸表情识别的多分辨率特征融合CNN[J].激光与光电子学进展,2018,55(7):071503.
[7] Ioffe S,Szegedy C.Batch Normalization:Accelerating Deep Network Training by Reducing Internal Covariate Shift[C]//32nd International Conference on Machine Learning,ICML 2015Lile,France:ICML,2015:448-456.
[8] He K,Zhang X,Ren S,et al.Deep residual learning for image recognition[C]//Proceedings of the IEEE conference on computer vision and pattern recognition.Las Vegas,Nevada:IEEE,2016:770-778.
[9] Huang G,Liu Z,Weinberger K Q,et al.Densely connected convolutional networks[C]//Proceedings of the IEEE conference on computer vision and pattern recognition.Honolulu,Hawaii:IEEE,2017.
[10] 张婷,李玉鑑,胡海鹤,等.基于跨连卷积神经网络的性别分类模型[J].自动化学报,2016,42(6):858-865.
[11] Lucey P,Cohn J F,Kanade T,et al.The extended cohn-kanade dataset(ck+):A complete dataset for action unit and emotion-specified expression[C]//Computer Vision and Pattern Recognition Workshops(CVPRW),2010 IEEE Computer Society Conference on.IEEE,2010:94-101.