Facial Expression Recognition Based on CNN-LSTM

Facial expressions are the most informative part of human communication, and the recognition of facial expressions by computers has become a hot topic in the field of computer vision. Before the development of deep learning, manual determination of models was relied upon for facial expression classification. With the advancement of computer technology, CNN has gradually become the optimal model for facial expression recognition. However, since CNN only models local features and does not consider the temporal information contained in facial expression sequences, overfitting is prone to occur during the modeling process. In this paper, based on the above situation, LSTM, which performs well in handling sequential data, is introduced to construct a CNN-LSTM fusion model. CNN is used to capture local features, and LSTM is utilized to capture global features for optimizing CNN. To validate the effectiveness of the proposed model, experiments were conducted on the CK+48 dataset, which contains 750 images. Firstly, CNN was optimized through hyperparameter tuning, and after determining the optimal learning rate, CNN was used for classification prediction, with LSTM added as a supplement to CNN. Finally, it was found that severe overfitting occurred with a single CNN, but after adding LSTM, the accuracy on the test set increased to 84%, effectively compensating for the shortcomings of CNN.