Face Expression Recognition using Side Length Features Induced by Landmark Triangulation
Avishek Nandi, Dutta Paramartha, Nasir Md · 2020
The human face is capable of generating six basic types of facial expressions: Anger, Fear, Disgust, Sadness, Happiness, and Surprise. Each expression is reflected in the face as deformation and dislocation of facial components of a given face image. The Active Appearance Model (AAM) is a statistical model of object shape recognition used to generate landmark points describing the geometrical position of the facial components. Landmark points generated by AAM are used to formulate triangulation description of the shape of the face. In this article, the authors have proposed a novel method using a different ratio of side lengths of a triangle constituting geometrical shape derived through forming triangulation generated out of landmark points indicating axial positions of facial 54 components fitted by an AAM. These novel side length features are learned by a MultiLayered Perceptron (MLP) network to segregate facial expression into six basic classes of expressions. The results obtained henceforth by applying the proposed method in JAFFE, Extended Cohn-Kanade (CK+), MUG and MMI databases are quite impressive.