AFIS stands for Automated Fingerprint Identification System, and it is a biometrics system for automated fingerprint identification; the system consists of a database of fingerprint images used for criminal identification purposes, and is used worldwide.
A very simple definition of a fingerprint impression is any impression of, or left by, the friction ridge skin of a finger or thumb. Friction ridges are the ridges and furrows that are on the pads of the fingers and thumbs. These ridges are there to provide grip, hence the name friction ridges. A palm print is the same thing, it’s just from the palm instead of the pads of the finger.
Every fingerprint will have characteristics that make it unique, and the characteristics are termed minutiae. Minutiae, are the major features of fingerprints that are used in matching one print to another.
Historically, fingerprint records required huge storage files, and lots of folks to maintain and search the thousands, or millions, of paper fingerprint cards the agency possessed.
Digital fingerprint databases are now the norm; they digitally store the rolled fingerprint impressions from each finger (“tenprints”) and fingerprint impressions of the fingers extended in what we call flat impressions. (For the purpose of this article, Latent Fingerprint Impressions, are fingerprint impressions recovered from a crime scene, that have not been identified as belonging to any certain person.)
With AFIS, police can enter a recovered latent fingerprint image into the system, allowing the recovered prints to be compared against a database of known fingerprint records. Additionally, there is a “latent to latent” function, that allows police to compare an unknown latent print against other unknown latent fingerprint, thus enabling officers to link crimes together. AFIS allows law enforcement to also Search a set of known fingerprints (tenprints) against the records of an existing tenprint record. Lastly, it lets officers search a new tenprint (from a new arrest record) against the record of existing unsolved latent prints saved in the system, typically referred to as an unsolved latent database.
Click here to see how latent fingerprints are processed – Identifying latent fingerprints; a basic look at how police match fingerprints from a crime scene.)
In 1980, NEC received a U.S. patent for automatic minutiae
detection system, and they began marketing the automated fingerprint identification systems to the United States a few years later. AFIS became commercialized, and it developed into a quick and reliable system to find potential matches in huge fingerprint databases.
As far as AFIS goes, there isn’t just one system, there are several companies who develop and market AFIS systems. NEC Corporation has one of the best AFIS platforms in my opinion, and it’s the system I have the most experience with. The NEC system was the first to use ridge counts and relationships between minutiae in its algorithm which allows its system to match distorted prints while maintaining matching selectivity. In addition, NEC’s use of “zone” data reduces the number of false minutiae by ensuring that only minutiae in the “clear zone” are used in a search. (the green areas in the example below represent excluded zones, they will be ignored by the computer – with this system, you can exclude a distorted area on the impression by excluding that area) .
Where NEC is what I am accustomed to, and most comfortable with, all AFIS systems are going to work on the same algorithm principle. Modern AFIS system algorithms will include minutiae points, the relationship of each individual point in relation to other points, and the skeleton or ridge flow of the print.
On the task of the search itself, the tenprint images are already in the database, and the latent to be searched will me scanned into the system, edited, search parameters set, then all is submitted by a qualified technician or officer. Once scanned in, the operator will first have to orient the print to be searched, both vertically and horizontally; a hard task on a partial print, and even harder on a partial Palmprint. Next, some programs will automatically mark minutia points, the most common being ridge endings and ridge bifurcations, if not, the operator will have to mark them manually. If the computer plots the points automatically, the operator should still edit the marked points, I wouldn’t enter a latent without editing it, because as good a the program is, it’s still not going to take the place of a human in the law enforcement setting because the computer doesn’t actually compare fingerprint images; it’s only conducting a binary code search, based on its algorithms, and the image itself is irrelevant to the search.
Some systems also allow the operator to exclude distorted parts of the print by “zoning out” poor quality areas, this will assist in a more productive search.
After the edits, Its time to set search parameters. Systems give the operator a wide range of search options, so searches can be customized for a better chance at a positive outcome. An example here is an inquiry of the inked impression from the right thumb of a person believed to be lying about their name, in this case, search parameters would better be set to check the unknown impressions against only “right thumbs” in the database. Here, the system would be searching the data base of right thumbs only, as apposed to all ten fingers times a million records, or 10 million images. Another parameter search would be to search a group of fingers, instead of all ten of them. An experienced tech or officer can sometimes make an educated guess about what fingers produced a certain latent fingerprint, thus making an initial search against only those fingers. An example here is that if the latent print is a loop pattern, that slants to the left, it is most likely a finger on the right hand. This is because fingerprints that are in a loop pattern, open to the little finger side of the hand about 97 percent of the time. Therefore, running this latent against only the right hand (fingers 1,2,3,4,5) speeds up the inquiry.
Today’s AFIS can often return a search of a million records in under a minute, but user volume in peak use times, and how the search is submitted, really determines the speed of the inquiry return. Searches may take several hours during volume peaks, and the way the query is run also makes a difference on the return time. Lastly, there are priority levels that can be adjusted on the query, a murder that just occurred could be bumped up in priority over all other inquires in the queue. Having said all that, a human, counting one fingerprint card per second, 24/7, with no breaks at all; would complete the count to one million in 11 days, 13 hours, 46 minutes, and 40 seconds, and that’s just counting them, nothing else.
Most latent impressions are only partial sections of the entire finger pad surface, and their quality is usually less than ideal, so a positive return rate of around 30 percent is not unheard of. And not to only blame the latent prints, many submitted fingerprint arrest records are poor quality.
The systems themselves have a higher than 99 percent accuracy rate, however not every inquiry has a matching record in the database and there will always be a high number of poor tenprint images on file, meaning a tenprint to tenprint inquiry will likely return an 80 percent success rate in my experience. (Just because an image is in the database, doesn’t mean it’s very good, it can still lack quality, hurting chances of a “hit”)
A quality latent print entered, where a quality tenprint match is in the database, will rarely record a miss; the proper candidate may not be the top candidate, but it’s usually in the top three.
Once the search is completed, the system will give the operator a list of “candidates”. The list will be assigned a numerical number, for instance, NEC assigns numbers that increase in size as the mathematical possibility of a match increases, with 9999 being the highest score possible. A score of 9999 will only be returned if a very detail rich and clear latent was entered, and a very good tenprint record was in the database.
There isn’t an average score for hits in my experience, but a candidate list that shows all having similar values are typically not ‘hits’, say all candidates having a score the 1700’s. Conversely, a number that is very much different from the crowd usually is indicative of a hit; say #’s 10 through #2 all have values in the 1700’s, then a 3600 is in the number one spot. Lastly, enter a great latent and get the top score of say 300, and you can bet just about anything that it’s match is not in the database.
These returned candidate images, regardless of the scores, will need to be examined by a technician or officer. The examiner will then make the final examination, and decision of the outcome personally.
A computer makes the work amazing efficient, however, a computer will never be the final authority on a match or a no match in a criminal case, it always takes a qualified person to make that call.
Click here for more about AFIS. https://www.thalesgroup.com/en/markets/digital-identity-and-security/government/biometrics/afis-history