Multiple classifiers fusion for facial expression recognition

Human facial expression recognition has been treated as a multi-class classification problem in the field of artificial intelligence. The main difficulty lies in how to distinguish the different categories of expression features. In this paper, we identify common facial expressions by fusing multiple weak classifiers. It compensates for the disadvantage of single classifier in weak generalization ability and low recognition rate for different datasets and different environments. This paper integrates the prediction results of each classifier through improved weighted mean value method and proposes an expression feature extraction method based on keypoint detection. Classifier fusion methods enable each classifier to perform at its best in order to improve overall expression recognition. Keypoint detection is used to improve the model’s attention on the expression features. Convolution neural network is selected as the model for feature extraction and classification, and the model structure is adjusted. Experiments show that the recognition accuracy of this method used on datasets FER 2013 and CK+ are 70.7% and 95.4% respectively, which are better than that of a single classifier, which shows that the keypoint extraction feature and classifier fusion method used in this paper have a good effect on facial expression recognition.