Reliable face recognition by random subspace support vector machine ensemble

Though many progresses have been made, face recognition is still a challenging topic in computer vision. Most of the published works focused on accurate classifiers design to produce identity predictions for query faces without suggesting how reliable the predictions are. These classifiers may not be applicable in some critical situations where the incorrect predictions have serious consequences. Aiming to tackle this problem, this paper proposes a highly reliable face recognition scheme by Random Subspace Support Vector Machine (SVM) ensemble which provides a reject option. Being different with previous classifier ensembles which purpose to increase the classification accuracy only, the objective of the proposed SVM ensemble is to supply classification confidence to accommodate the situations where no decision should be made if the confidence is low. The ensemble is created using Random Subspace (RS) method, together with four different feature descriptions to comprehensively characterize face images, including Local Binary Pattern (LBP), Pyramid Histogram of Oriented Gradient (PHOG), Gabor filtering and wavelet transform. The consensus from the ensemble's voting conforms to the confidence measure and the rejection option is accomplished accordingly when the confidence falls below a threshold. The reliable recognition scheme is empirically evaluated using a realistic face database created by the author, showing that pre-defined 100% accuracy can be reached with a rejection rate 7%.