Time-consuming process
The FBI distinguishes fingerprints by comparing the coordinates of the fingertips' ridge ending. Investigators file fingerprints based on basic physical characteristics (shape, swirl, etc.). This is a time-consuming and highly inaccurate process that can result in false matches, said Charles Wilson, a scientist at NIST.
In the 30 million fingerprint samples processed to date, the optically enhanced system has not indicated a single incorrect match. Such accuracy would make a computer security system nearly impossible to fool, he said. Also, the preliminary tests show a pattern-recognition accuracy of 100 percent, virtually eliminating the possibility of mistaken denials of access, Wilson said.
In optical pattern recognition, scientists load a fingerprint image onto a liquid crystal spatial light modulator (see figure). A reference beam interferes with the resulting Fourier spectrum to record a Fourier-transform hologram into the neural network. The system can then compare the image to any new fingerprint pattern introduced.
The optical system can more accurately enter the fingerprint pat-tern into the neural network system, which in turn can "recognize" and differentiate a pattern when it is reintroduced into the system.
Some possible commercial applications include the use of fingerprint images for credit card verification, access to automatic teller machines and access to the Internet in place of or along with passwords.
The laboratory is seeking corporate collaborators to move into the project's next phase: refining the neural network and testing the prototype system in industry. NIST plans to work with the Financial Services Technology Consortium, an organization of banks, financial service providers, technology companies, national laboratories, universities and government agencies.
However, Wilson estimates that complete commercialization is still a long way off. "It won't be ready for the next generation of end users," he said. "Maybe the one after that."