Facial expression recognition using deep neural networks

We develop a technique using deep neural network for human facial expression recognition. Images of human faces are preprocessed with photometric normalization and histogram manipulation to remove illumination variance. Facial features are then extracted by convolving each preprocessed image with 40 Gabor filters. Kernel PCA is applied to features before feeding them into the deep neural network that consists of 1 input layer, 2 hidden layers and a softmax classifier. The deep network is trained using greedy layer-wise strategy. We use the Extended Cohn-Kanade Dataset for training and testing. Recognition tests are performed on six basic expressions (i.e. surprise, fear, disgust, anger, happiness, sadness). To test the robustness of the classification system further, and for benchmark comparison, we add a seventh emotion, namely “contempt”, for additional recognition tests. We construct confusion matrix to evaluate the performance of the deep network. It is demonstrated that the network generalizes to new images fairly successfully with an average recognition rate of 96.8% for six emotions and 91.7% for seven emotions. In comparison with shallower neural networks and SVM methods, the proposed deep network method can provide better recognition performance.