Facial Emotion Recognition: A Comparison of Different Landmark-Based Classifiers

As stated by Ekman in his Facial Action Coding System (FACS), facial expressions can be interpreted as the activation of different sets of facial muscles. This recognition skill, however, demands extensive training and is indeed time consuming. Consequently, there have been many attempts to automate this process. In this paper, after applying common face detection and alignment algorithms to the Cohn-Kanade dataset, we fed a group of emotion-labeled landmarks to different classifiers in order to compare their results. The multilayer perceptron classifier showed the best performance with an average accuracy of 89%.