Emotion recognition from geometric fuzzy membership functions

The posterity challenging task in the Machine Intelligence field is to design a smarter system to identify the human emotions. Facial Emotion Recognition (FER) is a significant visual based tool to construct a smarter system that can recognize human emotions. The existing methods in FER are based on Action Units (AU), appearance and geometrical parameters. Nearly 7000 different combinations of AUs are used in AU to discriminate the emotions, which can be very expensive and increase processing time. Generalize appearance features across the universe is another challenging task. In this paper, a novel geometrical fuzzy based approach is presented to accurately recognize the emotions. The four corner features from eyes and mouth regions are extracted without considering reference face. The extracted features are used to define the quadrilateral shape that failed to match with the shapes in geometry. The degree of impreciseness exists in the quadrilateral is measured by the proposed Mixed Quadratic Shape Model (MQSM) using fuzzy membership functions. Finally, twelve fuzzy features are extracted from the membership functions and used by the classifier for validation. The CK, JAFFE and ISED datasets are used in the experiment to evaluate the performance of the MSQM. It is observed the proposed method performed better than the contemporary methods using twelve fuzzy features without reference image.