Fingerprint matching was developed in the 19th century. Although crude at the time, it opened our eyes to a new type of identification of individuals since each person has a unique set of fingerprints. Since its first development, we have discovered several ways to use this cutting edge technology. Forensic investigators use it regularly to identify criminals. The AFIS (automated fingerprint identification system) is being used more and more for establishing positive personal identification in applications like voter registration, driver's licensing, welfare, fraud prevention, and similar programs.
Although both of these uses for fingerprint matching use the same underlying technology, they are used in such different environments and for such different goals that one cannot simply use a forensic fingerprint identification system in the civil arena. Even in strictly the civil arena each identification system must be tailored to suit the application for which it is being used. However, even with all these differences, one factor remains common to all systems and that is a need for accuracy and speed. Forensic investigators have to be absolutely certain that they have the right person when they get a match and they need one as soon as possible so the criminal doesn't escape as they are waiting for a fingerprint match. For the AFIS, people want to make sure that their lives are secure, and no one can pretend to be them. They do not want to wait forever to be verified that they are who they are in order to continue on in the process of voter registration, getting a driver's license or which ever process they may be going through.
Accuracy and speed were two of our goals as we began this project. We knew how to do correlation-based matching with convolution but also knew this would take a very long time and was not always very accurate. So, we began searching for a new angle and decided to try type classification before the correlation algorithm and make our correlation algorithm faster by using the fft. This way we save time by running the algorithm on less potential matches, and we have less of a chance to have an incorrect match.