Real-Time Facial Expression Recognition Using Deep Convolutional Neural Network

Human-computer interaction (HCI) technology has constantly changed the life of mankind, and people's demands for interaction rise as well. From simple implementations of the instructions to the emotional interaction, a big step into the era of artificial intelligence has been made in the technology of interaction. The trend of HCI must be from mechanized computer instructions to natural language. Since more than half of the information in human's communication is contained in non-verbal factors, facial expression recognition becomes an indispensable part. In this paper, convolution neural network is used to learn the facial expression. We used two different data sets, one large but with low resolution, the other small but with high resolution. By means of the local directional number mode operator, preliminary features are extracted from the latter. For data set with low resolution, we only perform denoising and normalization to maintain the few residual features. Then a self-built data set is used to test these two models, and their generalization performance is compared by confusion matrix. Also a detailed discussion of the recognition accuracy of different expressions is given by comparison and analysis. Finally we got the accuracy of 59% and 65% respectively, and came to the conclusion that the expression Happy and Surprise own the most recognizable features, while the others are somehow similar to each other so special enhancement or classification method that focuses on the similarity should be carried out in the future.