Performance Analysis of Robust 2D-PCA and 2D-LDA Feature Descriptors on SVM

Abstract

Facialexpressions are the most intuitive way of communicating non-verbal messages. This type of communication provides effective response and feedback from the speaker to listener and vice-versa. In this paper robust macro facial expression recognition techniques are presented. 2D-PCA and 2D-LDA are robust geometric feature descriptors presented in this paper capable of cancelling noise and extracting maximum spatial features from image samples with unstable illumination condition which leads to correct classification of results. Experiments are carried out separately on both feature descriptors using Cohn Kanade (CK+) dataset, MMI dataset, and Japanese Female Facial Expressions (JAFFE) database for analysis. The experimental results are evaluated and their performance compared using Support vector machine (SVM) classifier with three kernels; linear, polynomial and radial basis function (RBF). The results from the proposed methods are also compared with existing novel methods and it is found out that the results from the proposed methods perform significantly well.