Robust multi-modal and multi-unit feature level fusion of face and iris biometrics