Subject Independent Facial Expression Recognition: Cross-Connection and Spatial Pyramid Pooling Convolutional Neural Network

Facial expression recognition is still a problem at present, especially in the case of individual independence. On the one hand, due to the influence of morphological changes, ethnic differences and other factors, the expression of individual expressions varies greatly. On the other hand, there is currently no publicly available large-scale dataset that can support deep neural networks. To this end, this paper proposes cross-connection and spatial pyramid pooling convolutional neural network. The model not only uses spatial pyramid pooling for high-level feature enhancement, but also combines cross-connection and spatial pyramid pooling to extract important low-level features. Finally the different levels of features are connected to improve the generalization performance of the model. We validate our approach in four widely used public expression datasets (CK+, JAFFE, MMI, NimStim). Compared to other facial expression recognition methods, our proposed method achieves comparable or superior results. In the case of subject independence, the model achieved a good result with 97.41% accuracy on the CK+ dataset.