Emotion Recognition on Selected Facial Landmarks Using Supervised Learning Algorithms

© 2018 IEEE. Facial landmarks may be used to localize the movement of facial muscles that help identify an emotion. It is important that these points are appropriately represented to achieve a successful emotion Recognition rate. In this paper, the extraction of 68 facial landmarks, normalization methods and classification of 7 basic emotions are presented. The Cohn-Kanade Database is used as a test bed for the different emotion Recognition tasks. The images are normalized by transforming the inputs based on similarity (CKCT) and the mean shape (CKMS). Forward Search and Principal Component Analysis are used to identify the most important features among the 68 facial points. Decision Tree, Logistic Regression, K-Nearest Neighbor and Multilayer Perceptron algorithms are used in building classifiers on reduced and complete feature set. It is interesting to note that facial points in the mouth area are found to be significant in the classification of emotions