A novel approach to FER using PCA, gabor wavelets and LBP

Abstract

FER has become a hot area of research due to its inevitable role in the fields of human computer interaction, pain detection in patients, automated learning systems etc. A new FER model is proposed in this paper. First, the dimension of the input images is reduced using PCA. Then features are extracted using either Gabor wavelets or LBP. SVM classifier is used for classification. The model is evaluated on benchmark datasets such as JAFFE, Cohn-Kanade and MMI. It has been observed that this model outperforms some of the earlier ones on facial expression recognition found in literature.