Abstract

In order to provide safety and security from fraudulent acts, it is necessary to use a reliable biometric identifier. Fingerprint is considered to be one of most effective biometric identifiers because of its universal characteristics. The recognition rate of identification/verification systems depends to a great extent on the quality of the fingerprint image. In a fingerprint recognition system, there are two main phases: 1) extraction of suitable features of fingerprints, and 2) fingerprint matching using those extracted features to find the correspondence and similarity between the fingerprint images. The low quality of fingerprint images provides false minutiae at the stage of feature extraction and reduces the recognition rate of minutiae-based fingerprint matching systems. Use of enhanced fingerprint images improves the recognition rate but at the expense of a substantially increased complexity. The objective of this research is to develop an efficient and cost-effective scheme for enhancing fingerprint images that can improve minutiae extraction rate as well as effectively improve the recognition rate of a minutiae-based fingerprint matching system.
In the first part of this thesis, a novel low-complexity three-stage scheme for the enhancement of fingerprint images is developed. In the first stage of the scheme, a linear diffusion filter driven by an orientation field is designed to enhance the low-quality fingerprint image. The computational complexity is reduced by using a simple gradient-based method for estimating the orientation field and by using a small number of iterations. Although some of the broken ridges in the fingerprint image are partially connected after the first stage, this stage has a limitation of not being able to connect ridges broken with wide creases, and also not being able to recover ridges in the smeared regions. To overcome the shortcomings of the first stage, the fingerprint image obtained after the first-stage enhancement is passed through a compensation filter in the second stage. Although the broken ridges in the enhanced fingerprint image after the second stage are fully connected, the ridges affected by smears are only partially recovered. Hence, the output obtained from the second stage is passed through the third-stage enhancement, which has two phases: short-time Fourier transform (STFT) analysis and enhancement by an angular filter. In the first phase, a Gaussian spectral window is used in order to perform the STFT and this window helps to reduce the blocking effect in the enhanced image. In the second phase, the image obtained from the STFT is passed through an angular filter, which significantly improves the overall quality of the fingerprint image.
In the second part of this thesis, the effectiveness and usefulness of the proposed enhancement scheme are examined in fingerprint feature extraction and matching for fingerprint recognition applications. For this purpose, a minutiae extraction algorithm is first applied to extract minutiae from fingerprint images and then a minutia-based matching algorithm is applied to the set of extracted minutiae using a hybrid shape and orientation descriptor in order to find similarity between a pair of fingerprints.
Extensive experiments are conducted throughout this thesis using a number of challenging benchmark databases chosen from FVC2000, FVC2002 and FVC2004. Simulation results demonstrate not only the effectiveness of the proposed enhancement scheme in improving the subjective and objective qualities of fingerprint images, but also a superior minutiae extraction rate and a recognition accuracy of the fingerprint images enhanced by the proposed scheme at a reduced computational complexity.