An Evolving Neural Network for Authentic Emotion Classification

Nowadays, there are few international databases based on authentic gesture. Most of the facial expression databases are not naturally linked to the emotional state of the test subjects. In this work, we expand the authentic emotion database created in 2003 by adding more subjects. Meanwhile we combine evolutionary algorithms with neural networks and well improve the recognition rate. We also implement other classification methods like gene expression programming and decision trees in order to compare with the adjusted neural networks. The experiment results show that our way to evolve back propagation neural network is quick and it can achieve an average recognition rate of 97%. Besides, it is much faster and more accurate than the gene expression commercial software: GeneXproTools, which is usually very powerful in many common datasets´ classification.