Evaluation of Feature Extraction Methods for Face Recognition

Abstract

Feature operators can transform raw pixel values of an image into a representation better suited to the later processing and classification steps in the face recognition system. In this paper, we evaluate the performance of 6 feature extraction methods, i.e., Local Binary Patterns, Histograms of Oriented Gradients, Scale Invariant Feature Transform, Speed-Up Robust Features, Fully Affine SIFT and Gabor features. Each feature was tested on 3 face databases of Yale, ORL and UMIST. The experimental recognition rate and matching time are given and compared to indicate different preferential features for different application conditions. ASIFT has the best result in recognition rate while SURF outperforms others in matching time.