Research Of Facial Expression Recognition Algorithm Based On Deep Learning

As an important means of intelligent human computer interaction, the technology of facial expression recognition has broad application prospect. The technology has been widely used in intelligent transportation, auxiliary medical treatment, distance education, interactive games, public safety and so on. In recent years, the technology of facial expression recognition has received more attention, and has become a hot spot in the field of computer vision and machine learning. Therefore, the research on facial expression recognition technology has important theoretical significance and practical application value.In this thesis, the effective feature extraction of facial expression is used as the main research object. The traditional method needs to be characterized by manual extraction and the general convolutional neural network lacks enough feature extraction ability, which limits their feature representation and generalization performance. To improve the feature extraction capability, this thesis presents a deep convolutional neural network model based on consecutive convolution. The core of the model is using small-scaled convolutional kernels to more detailed extract local features, and with the help of two consecutive convolutional layers increase non-linear expression ability of the model. Namely, the model receives the 2D face image, which is segmented by the face, then extracts the feature of face images by using two consecutive convolutional layer with 3×3 convolutional kernels, and the model can extract the local features better. In the convolution process, the ReLU activation function is used to replace the traditional sigmoid and tanh activation function, to increase the training speed of the model, and to solve the problem of gradient disappearance. Then, the max-pooling with 3×3 kernels is used as a pooling method, which can not only reduce the dimension of the feature, increase the computational performance of the network, but also highlights the details and makes the characteristics have a certain translation invariance. After the introduction of dropout technology to solve the problem of over-fitting. Finally, using two fully connected layers to classify the facial expression.By comparing the experiment and analysis in JAFFE and CK+ database, this method has obvious advantages in both the recognition performance and the generalization performance over the manual feature extraction method and the 2-layer and 3-layer convolutional structure. In the same experimental conditions with the existing literature, the model achieves a recognition rate of 100% for the JAFFE and the CK+ facial expression image database. It has a good performance and practicality in the analysis and recognition of facial expression. In this thesis, taking advantage of the designed facial expression recognition system, the facial expression images batch pretreatment, training and testing the deep consecutive convolutional neural network, and other functions are basically achieved, and facial expression recognition is completed by means of the good trained model successfully.