Combined Optical and Neural Network Fingerprint Matching

摘要

This paper presents results on direct optical matching of inked and real-timefingerprint images. Direct optical correlations and hybrid optical neural network correction are used in the matching system for inked fingerprints. Preliminary results on optical matching of real-time fingerprints use optical correlation. The test samples used in the inked image experiments are the fingerprints taken from NIST database SD-9. These images, in both binary and gray level forms, are stored in a VanderLugt correlator (1). Tests of typical cross correlations and auto correlation sensitivity for both binary and 8 bit gray images are presented. When global correlations are tested on a second inked image, results are found to strongly influence by plastic distortion of the finger. When the correlations are used to generate features that are localized to parts of each fingerprint and combined using a neural network classification network and separate class-by-class matching networks, 84.3% matching accuracy is obtained on a test set of 100,000 images pairs. Initial results with real time images suggest that the difficulties resulting from finger deformation can be avoided by combining many different distorted images when the hologram is constructed in the correlator. Testing this process will require analysis of 10 to 20 seconds sequences of digital video.