Biometrics has many problems, on of them is with intra-class variations (between the same class). For example more photos of the same individual with different face expressions or angles from where the photo is taken. This problem raises inside of an uncontrolled system.
Another problem is with very small inter-class variations and inter-class similarities (between different classes). An example is twins or two very similar individuals. This can obviously cause some problems for face recognition. The pose, the illuminations and the other factors also play an important role in the similarity of two already similar individuals.
Another problem is nosiy and/or distorted acquisitions. For example a dirty sensor, poor quality fingerprints (heavy workers), non uniform lighting. A fingerprint sensor my contain some remains of the previous fingerprint (of another individual) and this can cause some noise in the current fingerprint sample.
This problems are summarized by an anocrym by the name of PIE (Pose, Illumination and Expression), in the future transformed in APIE (Age, PIE).
Non-universality is another problem. For example blind people may have such a deep illnes to the eyes that it’s impossible to capture the iris. Another example is voice recognition for mute people.
Failure to enroll
A failure to enroll happens when an user cannot be enrolled because of the too low quality of the sample. In this case it’s useful to have an human intervention, maybe on the condition of maybe the system can ask the user to input the sample again.