Automatic Facial Expression Recognition Using Histogram Oriented Gradients (HoG) of Shape Information Matrix

Abstract

Automatic Recognition of Facial Expression (AFRE) has many potential applications in Human–Computer Interaction (HCI), security and surveillance, smart emotion-driven tutoring, etc. Human beings often display their emotions in various media of communication like language, voice, gesture, and facial expression. The contribution of facial expression is maximum among other means of communication. We have proposed a novel shape-oriented representation of facial structure using appearance information of the face. Active Appearance Model is used to select 25 salient landmark points on the face and a Shape Information Matrix (SIM) is formed by forming triangles on the face using landmark points. Histogram oriented Gradients feature of the Shape Information Matrix (SIM) is used to extract local orientation features of facial image. The proposed machine is trained using MultiLayer Perceptron (MLP) Neural Network and well tested over CK+, JAFFE, MMI, and MUG benchmark databases. Results obtained are impressive and encouraging.