An Expression invariant Face Recognition based on Proximal Support Vector Machine

Face recognition with variations in expressions is a challenging task as various facial structures are altered. In this work, an expression invariant face recognition system is proposed in eigenspace based on proximal support vector machine (PSVM). PSVM is an enhanced version of support vector machine (SVM), where a system of linear equations is applied for generation of linear and non-linear classifiers. This makes classification process faster due to low computational complexity as compared to SVM. In proposed method, training and test images are first processed using principal component analysis (PCA) to reduce the feature set, that needs to be accounted by PSVM for multi-class classification. A remarkable 100% and 98.33% accuracy was achieved on two benchmark face databases JAFFE and Yale respectively. This validates the fact that face recognition based on PSVM is invariant to facial expressions while being almost ten times faster than its traditional counterpart, SVM.