Authentication based on face recognition under uncontrolled conditions

This paper proposes a method to address issues regarding uncontrolled conditions in face recognition. This method extracts affecting factor from the test sample utilizing mask projection. Current methods remove occlusion from test sample and reconstruct it. Unlike these methods, proposed method tries to add extracted occlusion to all normal training samples and compares test sample with all synthetic affected training samples. The method has been applied for multi-factor authentication/verification based on face biometric. Obtained results indicate high accuracy, comparable to the best sparse method, in the lake of sufficient training samples for each class(single sample classes).