A Comparative analysis of appearance based subspace methods for facial expression recognition
Silma Shama Broti · 2019
Facial expression recognition is like a pattern recognition problem, which is divided into two parts dimensionality reduction and classification. It is very much challenging to find out an efficient combination of dimensionality reduction technique and classifier. Many appearance based dimensionality reduction techniques have been proposed in recent years. This research focuses on the comparative analysis of appearance based subspace methods (PCA, KPCA, PCA+MI, KPCA+MI) for facial expression recognition in completely equal working conditions. PCA is a vastly used feature extraction technique which is linear and unsupervised. A kernelized version of PCA is called KPCA increases classification accuracy by handling non-linear data which is an unsupervised feature extraction method. KPCA shows the highest accuracy among these two methods. Combining a feature selection method MI with KPCA which can highly increase the recognition rate. The accuracy of KPCA+MI was above 95.24% when the moderate number of training features were used to train the SVM with RBF kernel.