Face expression recognition based on a two-branch convolutional neural network

Inhalt:
In order to further improve the accuracy of facial expression recognition, a deep convolutional neural network algorithm integrating global and local features is proposed. The algorithm consists of two improved convolutional neural network branches, global and local branches, which are used to extract global and local features, respectively, weighted fuse the features of the two branches and classify using the fused features. First, the global features are extracted using the improved VGG16 network model, the local features use the directional gradient histogram algorithm to obtain the local texture information of the original image and input it to the shallow convolutional neural network to automatically extract the local features related to the expressions. Again, two fully connected layers are used to assign different weights and weighted fusion. Finally, the softmax classifier is classified. The experiment was validated on the CK + and FER2013 datasets, and the classification accuracy reached 96.97% and 71.66%, respectively. Compared with several other algorithms, the algorithms generally performed well, have good recognition effect and good robustness, which can provide an effective basis for facial expression recognition.