A fingerprint recognition system for person identification

Abstract

Biometric traits such as fingerprints, face, iris and voice have been shown to be not only convenient to use but also highly secure person authentication metrics. In particular, they are more reliable than current knowledge - and token-based methods since several biometric modalities can uniquely define a particular individual. The fingerprint is one of the biometric traits characterised by a high amount of uniqueness, such that even the fingerprints of the same person are different. As a result, they constitute almost 70% of the biometrics market revenue. In this project, an automated system was developed to discriminate with a satisfactory degree of confidence between persons using their fingerprints by comparing multiple impressions in a database with each other and obtaining a similarity score for each pair of matched fingerprints. This was achieved by first pre-aligning the two fingerprint images to be compared using a 2D correlation function, followed by extraction of the local ridge orientation and local ridge frequency. These were then used for enhancement of the fingerprint image using a Gabor filter to reduce the effects of any noise that is typically present in the impressions. The resultant image is then binarised using Otsu's method and the ridges of the fingerprint are thinned so that extraction of minutiae can take place. Segmentation is then done so that any minutiae found in noisy regions or in the background are eliminated. Any erroneous minutiae are then filtered out so that the minutiae from the template and input images can be compared with each other in terms of their spatial and angular differences. A method to optionally classify fingerprints according to their global pattern was also implemented using the Poincare index in conjunction with a median filter. Encouraging results were obtained using a variety of fingerprint databases, achieving a classification error rate of 3.75% and an Equal Error Rate (EER) of 7.03%, which are comparable to fingerprint recognition systems found in the market.