Alternative Matching Schemes For Automatic Fingerprint Identification Systems

Biometrics is a recognition technology that uses the unique behavioral and physiological traits of the human body as identifiers. The fingerprint is one of the oldest and most widely used forms of biometric identifiers. In this thesis, an automatic fingerprint identification system designed by a previous Electrical Engineering graduate student at The University of Texas at Arlington is upgraded with the purpose of improving matching accuracy. A review of the baseline automatic fingerprint identification system is provided as a reference point for improvements in the system. Then, improvements to the image enhancement stage of the baseline system are discussed. They include replacing the simple transformation contrast stretching algorithm with the "contract and stretch" algorithm and upgrading the threshold variance in image segmentation to a threshold variance to mean ratio. Next, improvements to the feature extraction and fusion stage of the baseline system are covered. A new method for quantizing ridge direction estimates in the direction image is included in this section. Finally, alternative matching schemes are implemented in the correlation-based matching stage of the baseline system. The new matching schemes include direction only feature, density only feature, two channels with addition-type fusion, two channels with multiplication-type fusion and two templates per class. At the end, simulations are performed in order to test the performance of the improvements implemented in the automatic fingerprint identification system.