Comparative Study of Statistical Models and Classifiers in Face Recognition

In lieu of a robust face recognition system, fetters like orientation, pose and lighting are the foremost encounters to be addressed. Gabor wavelets are employed in the proposed method which has their biological relevance and computational properties similar to human eyes. To eliminate the variations due to pose, lighting, orientation and features to some extent, preprocessing of human face image is to be done in a well-organized manner. Principal Component Analysis (PCA) and Linear Discriminant analysis (LDA) are widely adopted prospective face recognition algorithm to extracts low dimensional and more discriminating features from face. The classification is done using Euclidean distance classifier, Mahalanobis Metric classifier, Artificial Neural Networks (ANN) and Support Vector Machine (SVM). For the proposed work the train dataset is considered in randomized fashion, means the database contains a set of images of an individual in which different set of train dataset is taken randomly to choose the trained set that gives high rate of recognition. This will support in dropping the overall database size and surge the delineation of the system. The system has been successfully tested on AT&T Database of faces, YALE database and our own AMTR database. From the exploration, it's evident that Gabor plays a major role to eradicate variations, PCA and LDA when glued together, extracts more judicious features than PCA features. Back-Propagation neural network (BPN) and SVM are bit complex classifiers but gives improved recognition when compared to distance measure methods. The overall system achieves better accuracy as the complexity grows up and gives a better recognition rate with more number of features.