Facial expression recognition using distance and texture signature relevant features

Abstract

Distance and texture characteristics among the landmark points reflected in human faces are important features in so far as the recognition of human faces is concerned. In this article we consider (i) normalized distance signature obtained from Active Appearance Model (AAM) based grid, (ii) normalized texture signature derived from salient landmarks within the grid, (iii) stability indices arising out of these signatures and (iv) relevant statistical measures as the set of features for training of artificial neural models such as Multilayer Perceptron (MLP), Radial Basis Function Network (RBF), Nonlinear AutoRegressive with eXogenous input (NARX) and Convolutional Neural Network (CNN) to achieve the task of recognition of facial expressions. The Cohn-Kanade (CK+), Japanese Female Facial Expression (JAFFE), MMI and MUG benchmark databases are used to conduct the experiments and the results obtained justify the effectiveness of the proposed procedure. The combined distance-texture (D-T) signature is found to perform convincingly better than the distance signature and texture signature individually. The effectiveness of the proposed technique based on combined D-T signature is substantiated by its extremely encouraging performance compared to other existing arts.