Reliable face recognition using adaptive and robust correlation filters

This paper expands on the scope of correlation filters to show their usefulness for reliable face recognition. Towards that end we propose adaptive and robust correlation filters (ARCF) and describe their usefulness for reliable face authentication using recognition-by-parts strategies. ARCF provide information that involves both appearance and location. The cluster and strength of the ARCF correlation peaks indicate the confidence of the face authentication made, if any. The development of ARCF, motivated by MACE filters and adaptive beam-forming from radar/sonar, is driven by Tikhonov regularization. The adaptive aspect of ARCF comes from their derivation using both training and test data, similar to transduction, while the robust aspect benefits from the correlation peak optimization to decrease their sensitivity to noise and distortions. The comparative advantages of ARCF are motivated, explained, and illustrated vis-ل-vis competing correlation filters. Experimental evidence shows the feasibility and reliability of ARCF vis-ل-vis occlusion, disguise, and illumination, expression, and temporal variability. The generalization ability of ARCF is further illustrated when decision-making thresholds learned a priori from one data base, e.g., FERET, carry over to face images from another data base, e.g., AR.