Facial Expression Recognition Based On The Ensemble Deep Learning Model

Emotion is the subjective feeling of human and animals to the external objective experience,and the expression is the external way which used to convey the emotional message.Facial expression,tone of the language expression and posture expression are the three main forms of expression.Facial expression refers to the transfer of various emotional states by controlling the muscles of face,eyes,corner of mouth and so on.It can not only convey the true feelings of people,but also promote the silent exchange of ideas between strangers.Therefore,the facial expression is a very important and effective non-verbal way in interpersonal communication.In the field of Human-Computer Interaction,the facial expression recognition has very important theoretic meaning and research value.If computer can understand the facial expression of human,then the computer will no longer be an icy machine for us and can react adaptively to the analysis results of facial expression.The technology of facial expression recognition endows the computer with the ability to understand human emotions and can promote the development of the intelligent home,smart city and interrelated industries.However,due to the inherent characteristic of facial expressions,there are major technology challenges for computers to recognize facial expressions through traditional approaches.It is very difficult for computers to extract the quantized features of facial expressions because the facial expressions are hard to represent by the quantified methods.Because of this,we cannot simply classify the facial expression recognition challenge as the image classification problem and should pay more attention to the extraction of the abstract features of facial expressions.Recently,with the development of deep learning and machine learning,various neutral networks have been proposed such as the Convolutional Neutral Networks(CNNs),Generative Adversarial Networks(GANs)and the Recurrent Neural Networks(RNNs)and those up-to-data networks can bring new ideas for researchers.In this paper,a facial expression recognition algorithm based on deep learning and ensemble learning is proposed.We adopted three CNNs-based models with different structures as the sub-networks and trained these sub-networks independently.According to the val-accuracy,each sub-network was assigned a weight.Simple voting and weighted voting strategies were applied to combine the three sub-networks into the final ensemble model.During the test phase,a new recognition strategy was used to further improve the recognition accuracy and the parallel computing method was used to improve the computational efficiency.Our proposed algorithm was evaluated on the Fer2013 dataset,Jaffe dataset and the AffectNet dataset.The experimental results show that our proposed model achieves a test accuracy of 71.91%,96.44% and 62.11% on above mentioned datasets,and increases the test accuracy by approximately 2-3% than unique sub-networks.The results of the experiments on the Fer2013,JAFFE and AffectNet datasets were ideal,indicating that our proposed facial expression recognition algorithm has a relatively good performance compared with other algorithms.