A perfect biometric recognition system would have a FRR = 0 and a FAR = 0 which is a little bit unachievable in reality. It is also interesting that any of the two values FRR and FAR can be reduced to an arbitrary small number, with the drawback of increasing the other value. Another interesting value is the Total Error Rate TER equation (3) which is defined as:
At this point it is important to emphasize the fact that these measures could be heavily biased by one or either type of errors (FAR or FRR) depending only on the number of accesses which have been used in obtaining these respective errors. This means that the TER will always be closer to that type of error which has been obtained with the largest number of accesses. The overall performance of a biometric recognition system should not be measured by the TER but rather by the Receiver Operation Characteristic ROC, which represents the FAR as a function of the FRR. So wherever there is a tradeoff of error types, a single performance number is inadequate to represent the capabilities of a system. Such a system has many operating points and is best represented by a performance curve. The ROC curve has been used for this purpose. Generally false alarm is plotted on the horizontal axis whereas the correct detection rate is plotted on the vertical axis.
In some applications though the Detection Error Tradeoff DET curve has been found to be more useful since both types of errors are plotted on the DET curve. Typically one can observe approximately straight lines, which do correspond to normal likelihood distributions, in DET plots. This method is especially useful in speech applications . Some high security applications tend to keep the FAR as small as possible when they operate at the point on the ROC. Forensic science operates with a very low FRR and a very high FAR since they desire to catch a criminal even at the expense of examining large numbers of false accepts. Civil applications try to work at a level where FRR and FAR are both as low as possible see Figure 3.
The recognition accuracy is depending on the image acquisition, the position of acquiring sensor, intensity of light focusing, environmental changes, noise, and bad user’s interaction with the sensor. Therefore the two images acquired by the sensor may not be having same characteristics. The biometric matching systems are used to find the matching score between the two images. The threshold t is assumed and the matching score is less than t then the image is considered as the different person. Then two errors are measured in terms of False Acceptance Rate FAR and False Rejection Rate FRR. FAR: The biometric measurement between two persons is same. FRR: The biometric measurement between two persons is different. If the system decreases t to make the system more tolerant to input variation and noise, FAR increases. On the other hand if the system increases t to make the system more secure, FRR increases accordingly. Figure 4 shows the performance of the system which is depending on the matching score between two images and is measured by the errors; False Acceptance Rate FAR and False Rejection Rate FRR.
Biometrics is an essential component of any identity-based system, but they themselves are vulnerable. Some of these attacks are simple to execute; solutions to these attacks have been identified, but there is still room for improvement. Attacks on biometric systems can result in loss of privacy and monetary damage, so the users need to be convinced about the system protection. New security issues with biometric systems may arise as their use becomes more widespread. In spite of this, biometric systems are being deployed for securing international borders, controlling access and eliminating identity theft.