Fingerprint Matching Algorithms
Fingerprint matching algorithms vary greatly in terms of Type I (false positive) and Type II (false negative) error rates. They also vary in terms of features such as image rotation invariance and independence from a reference point (usually, the "core", or center of the fingerprint pattern). The accuracy of the algorithm, print matching speed, robustness to poor image quality, and the characteristics noted above are critical elements of system performance.
Fingerprint matching has an enormous computational burden. Some larger AFIS vendors deploy custom hardware while others use software to attain matching speed and throughput. In general, it is desirable to have, at the least, a two stage search. The first stage will generally make use of global fingerprint characteristics while the second stage is the minutia matcher.
In any case, the search systems return results with some numerical measure of the probability of a match (a "score"). In tenprint searching, using a "search threshold" parameter to increase accuracy, there should seldom be more than a single candidate unless there are multiple records from the same candidate in the database. Many systems use a broader search in order to reduce the number of missed identifications, and these searches can return from one to ten possible matches. Latent to tenprint searching will frequently return many (often fifty or more) candidates because of limited and poor quality input data. The confirmation of system suggested candidates is usually performed by a technician in forensic systems. In recent years, though, "lights-out" or "auto-confirm" algorithms produce "identified" or "non-identified" responses without a human operator looking at the prints, provided the matching score is high enough. "Lights-out" or "auto-confirm" is often used in civil identification systems, and is increasingly used in criminal identification systems as well.
Read more about this topic: Automated Fingerprint Identification