Click here for Top of Page
Right Arrow: Next
Right Arrow: Previous

Fingerprints: Proving Ground for Pattern Recognition


By Anil Jain

(University Distinguished Professor, Michigan State University, USA)

Two fingerprints and a portrait photo have been captured for over 60 million visitors to the US through the US-VISIT program; 1,100 criminals have been denied entry.

Review by: 

Pranab Mohanty

University of South Florida


Anil K. Jain first enlightened the audience with a rich history of fingerprints and fingerprint matching systems. He illustrated the history of fingerprints with the quote from Galton in Nature, 1888—“Perhaps the most beautiful and characteristic of all superficial marks are the small furrows with intervening ridges and pores that are disposed in a singularly complex yet even order on the under surfaces of the hands and the feet”—and then highlighted the wide usage of fingerprint systems, such as AFIS (Automated Fingerprint Identification System) by law enforcement agencies over last 40 years.


Since a fingerprint is believed to be unique to each person, Jain emphasized that the biometric recognition market is expected to be dominated by reliable and highly accurate fingerprint based personal identification systems. With such a high expectation of matching accuracy and demand for an automated fingerprint matching system, Jain suggested that detailed and extensive research needs to be done to achieve the objective of a fully automatic, fingerprint-based person identification system.


In this talk, Jain pointed out various research areas for the scientists and engineers who are interested in designing fingerprint recognition systems. Some of these areas are sensor, spoof detection, extended feature set, uniqueness of fingerprints, multi-biometrics, template security etc. According to Jain a new set of fingerprint sensors such as 3D fingerprint imagery and high resolution sensors will have much impact on fingerprint matching technologies. Jain also provided some insight into the widely used minutiae based features for fingerprint matching and suggested that the extended set of level 3 feature sets, such as location of pores will  be helpful in detecting live fingers and will reduce the fake fingerprint threats. Jain also pointed out that, although fingerprints are assumed to be unique for each living person, a detailed test of this hypothesis is necessary and the error rate should be quantified. In the same context, Jain also raised his concerns on template security and highlighted his work on developing a secured database for fingerprint systems known as Fingerprint Fuzzy Vault system.


In future work, Jain is confident in developing automated fingerprint systems for latent matching, i.e., matching complete fingerprints to the partial fingerprints that commonly appear at crime scenes.


In conclusion to his talk, Jain provided various grounds of challenging tasks for developing a highly accurate fingerprint matching system. More on this talk can be found at Anil K Jain’s website,


Review by:

Daniel Zuwala

LORIA-INPL-Université de Nancy, France


Dr. Jain began with some background on fingerprints. Fingerprints are composed of ridges, that prevent slipping while grasping. There are different types of ridges that may be used for indexing: whorl, loop and arch. We know that the ridge formation starts at 1 or 2 focal points and spreads. Fingerprints are interesting because they are unique, permanent, can be classified, and denote genotype and phenotype characteristics.


Currently, the FBI has a database of 50 million fingerprints. Each day, about 50,000 searches are performed, each taking about 2 hours. Only 15% of the matches are reliable. There is a great demand for fingerprints recognition technologies in several domains including border security, financial fraud, and user convenience. These require inexpensive and compact sensor, and fully automated matching.


Local ridge characteristics or minutiae, ridge endings and bifurcations, can be extracted from the fingerprints. Singular points like core and delta can also be extracted. The matching is done by estimating the rotation, translation and distortion. Due to a large intra-class variation, the matching techniques are unlikely to be error-free. For instance, fingerprints collected after last year’s  bombing in Madrid led to the arrest of a man. However, the suspect was released after proving he was in the US when the bomging took place. This show us not to put complete faith in fingerprints. Current systems have a false reject rate of 0.1%, meaning that with 100,000 passengers per day at the Hong Kong airport, 100 "good" passengers would be stopped daily.


There are many research directions. The detection of artificial finger (spoof detection) is one of them.  One example of spoof detection is by multi-spectral imaging (MSI) sensors that can capture both the surface and the subsurface ridges, something difficult to construct on a spoof fingerprint. Another research direction is in minutia storage. Because spoof fingerprints can be reconstructed with minutiae stolen from a database, a secure way to save these data is needed. One very interesting research direction is in fingerprint capture technologies. Instead of using rolled inked impressions, a 3D image sensors can be used to collect fingerprints in a faster and more reliable manner.


To conclude, fingerprints are the earliest and the largest deployed application of pattern recognition. However, even today, fingerprint systems have non-zero error rates. More robust, accurate and cost-effective fingerprint matching systems are still needed.