Face Authentication Using Supervised Learning Techniques

The growing popularity of face biometrics for human authentication is due to its high user convenience and acceptance. Face recognition has evolved as an important application of biometrics and has been a topic of interest for several machine learning and computer vision communities. However, most of the attempts on face based personal authentication rely on decision threshold for accept or reject of the claimed identity. This paper investigates supervised learning techniques for face verification. The presented approach deals with computation of 5th level Haar wavelet coefficients of image used as feature for training of the classifiers like SVM, fuzzy SVM and KNN. The extracted biometric features are matched to compute genuine and impostor matching scores. The error rates FAR and FRR are then calculated using cross validation of the test set. The experiments are carried out on Yale database of 37 users with 25 images of each user. In our work we obtained FAR and FRR of 0.3285 and 0.1967 respectively which demonstrates the reliability of the proposed work.