The accuracy of face recognition system is often defined in terms of two parameters, False Rejection Rate (FRR) and False Acceptance Rate (FAR).

       FRR measures how often an authorized user, who should be granted access, is not recognized, while FAR measures how often non-authorized user, who should not be granted access, is falsely recognized.

       The control of FRR and FAR through recognition threshold adjustment defines the accuracy of face recognition system. When the recognition threshold is increased, it will adversely cause FAR to decrease. At the same time however, the increase of recognition threshold will result in the increase of FRR. For example, when recognition threshold is set at 100%, FRR rate will increase to its highest level.

        In accordance to the graph below, the curves created by FAR and FRR rates would intersect at a point called Equal Error Rate (EER). At this point, the errors resulting from combination FAR and FRR are at the lowest level achieved.      

       Setting the recognition threshold at this balance is the most effective. Thus EER is generally used as a standard setting for recognition threshold in face recognition system. The lower is the EER, the more accurate is the face recognition system.

       In its application however, the purpose and environment where face recognition system is used must be taken into consideration. For example, when such system is used for highly controlled environment, the FAR must be set at the lowest possible value so that there is lowest risk of intrusion by non-authorised users. On the other hand, authorized users may have higher level of rejection rates (a result of higher FRR).

       In the scenario where face recognition system is used for door control for office building, the traffic at the door must be taken into consideration, thus setting a higher FAR rate and lower FRR rate will be more efficient.