Geometric based Feature Extraction Techniques for Facial Expression Recognition using SVM

Abstract

In recent years, facial expression recognition becomes one of the interesting research topic in developing human computer interaction applications. In geometric-based methods, irrelevant features increases the computational cost of the facial expression recognition system (FERs). To overcome this issue, a geometric-based feature extraction method is proposed to extract the most relevant feature map to recognize six basic facial expressions including neutral. In this proposed method, the difference in measure of distance and triangle formation of 9 facial landmark points are calculated for each image with reference to the neutral image. Moreover, the subtle changes in facial muscle forms the deformation of facial points for different facial expressions. Hence, the difference of Euclidean distance and triangulation of each facial points of diverse expression images are calculated with respect to neutral image which are generated as feature maps and given to the SVM classifier. Further, the proposed method is evaluated using CK+, MMI and Self-collected datasets.