Genetic Programming-Based Feature Learning for Facial Expression Classification

Facia1 expression classification is an important but challenging task in artificial intelligence and computer vision. To effectively solve facial expression classification, it is necessary to detect/locate the face and extract features from the face. However, these two tasks are often conducted separately and manually in a traditional facial expression classification system. Genetic programming (GP) can automatically evolve solutions for a task without rich human intervention. However, very few GP-based methods have been specifically developed for facial expression classification. Therefore, this paper proposes a GP-based feature learning approach to facial expression classification. The proposed approach can automatically select small regions of a face and extract appearance features from the small regions. The experimental results on four different facial expression classification data sets show that the proposed approach achieves significantly better results in almost all the comparisons. To further show the effectiveness of the proposed approach, different numbers of training images are used in the experiments. The results indicate that the proposed approach achieves significantly better performance than any of the baseline methods using a small number of training images. Further analysis shows that the proposed approach not only selects informative regions of the face but also finds a good combination of various features to obtain a high classification accuracy.