August 10, 1998
Contact: Andrew Careaga Phone: 573-341-4328; [email protected]
RESEARCHERS DEVELOPING BETTER FINGERPRINT IDENTIFICATION METHOD
ROLLA, Mo. -- Researchers at the University of Missouri-Rolla are developing an intelligent computer system that will improve the efficiency of fingerprint identification methods used by governmental agencies, financial institutions and other high-security organizations.
The researchers in UMR's Smart Engineering Systems Laboratory (SESL) are using artificial neural networks rather than traditional methods to identify fingerprint characteristics. The newer method works better than traditional identification methods, such as searching for a duplicate fingerprint image in a large computer database. this is because neural networks are "trained" to recognize prints despite smears and smudges in the fingerprint images, the UMR researchers say.
"It is this flexibility and power that could lead to more reliable and affordable security systems," says Dr. Cihan H. Dagli, professor of engineering management at UMR and director of the SESL.
"Current fingerprint identification techniques require the development of a database containing multiple images of each fingerprint, and then a computer system matches a fingerprint with those in its database," says Dagli, who also is editor in chief of the Smart Engineering Systems Journal.
The traditional fingerprint recognition method involves the classification of fingerprints by ridge patterns. Prints with distinguishing loop patterns are placed into one category, while those with whorls are placed in another, Dagli says. When a person places his or her thumb into one of these identification systems, it first determines which category the print falls into -- looped, whorled or arched, for example -- and then sorts through the database of available prints in search of specific traits that will link that print to the individual.
These types of systems treat fingerprint identification and fingerprint classification as two separate processes. They require a lot of computing power and are not always reliable. A thumb smudge, for example, can cause such a system to incorrectly deny access to someone.
The system being developed at UMR would simplify the identification process, Dagli says. It involves the use of artificial neural networks -- computer systems that are modeled rudimentarily on the human brain and nervous system. Artificial neural networks are one of many "smart" systems now being used in manufacturing, medicine, robotics and other applications that can learn and solve problems by adapting to their environment, reprogramming themselves based on information they pick up from training and experience.
Using fingerprint images as input for this experimental identification system, the UMR researchers begin by removing the "noise" such as smudges and smears from the images prior to classification. They use a combination of computer programs to classify the images.
After categorizing prints according to their ridge characteristics, Dagli and his colleagues use another neural network to recognize the smaller distinguishing characteristics of fingerprints. This final step is accomplished through an "unsupervised" neural network that recognizes the specific coordinates of ridges and the direction of the tiny details of each fingerprint. "This classification and recognition stage provides optimum grouping of the individual fingerprints according to their ridge characteristics," says Dagli.
Early tests of the neural network solution are encouraging, Dagli says, because results are comparable to the success rates of existing systems. In tests on 28 fingerprints from six different people, the neural-network model achieved "confidence levels" of 0.97 to 0.99, with 1.0 being a perfect match.
-30-
RSS