Classification of facial expressions by using 2D empirical mode decomposition

In this study, we present a method to estimate six basic facial expressions of a person by applying 2-dimensional Empirical Mode Decomposition (2D-EMD) on face images which represent the emotional state. Three intrinsic mode functions (IMF) and a residue is obtained by applying 2D-EMD to each facial regions that are cropped from upper and lower face regions. Filters having different bandwidths are designed to detect the efficient frequency band of all eight sub-images. Two approaches are used to calculate the features: statistical metrics of each sub-image, and Radon Transform of sub-images are used as features in classification. Support Vector Machines classifier is used as to identify the facial expression. The results for the two feature vector approaches are presented and compared.