Fingerprint Recognition: A Histogram Analysis Based Fuzzy C-Means Multilevel Structural Approach

摘要

In order to fight identity fraud, the use of a reliable personal identifier has become a necessity. Fingerprints are considered one of the best biometric measurements and are used as a universal personal identifier. There are two main phases in the recognition of personal identity using fingerprints: 1) extraction of suitable features of fingerprints, and 2) fingerprint matching making use of the extracted features to find the correspondence and similarity between the fingerprint images. Use of global features in minutia-based fingerprint recognition schemes enhances their recognition capability but at the expense of a substantially increased complexity. The recognition accuracies of most of the fingerprint recognition schemes, which rely on some sort of crisp clustering of the fingerprint features, are adversely affected due to the problems associated with the behavioral and anatomical characteristics of the fingerprints. The objective of this research is to develop efficient and cost-effective techniques for fingerprint recognition, that can meet the challenges arising from using both the local and global features of the fingerprints as well as effectively deal with the problems resulting from the crisp clustering of the fingerprint features. To this end, the structural information of local and global features of fingerprints are used for their decomposition, representation and matching in a multilevel hierarchical framework. The problems associated with the crisp clustering of the fingerprint features are addressed by incorporating the ideas of fuzzy logic in developing the various stages of the proposed fingerprint recognition scheme.udIn the first part of this thesis, a novel low-complexity multilevel structural scheme for fingerprint recognition (MSFR) is proposed by first decomposing fingerprint images into regions based on crisp partitioning of some global features of the fingerprints. Then, multilevel feature vectors representing the structural information of the fingerprints are formulated by employing both the global and local features, and a fast multilevel matching algorithm using this representation is devised. udInspired by the ability of fuzzy-based clustering techniques in dealing more effectively with the natural patterns, in the second part of the thesis, a new fuzzy based clustering technique that can deal with the partitioning problem of the fingerprint having the behavioral and anatomical characteristics is proposed and then used to develop a fuzzy based multilevel structural fingerprint recognition scheme. First, a histogram analysis fuzzy c-means (HA-FCM) clustering technique is devised for the partitioning of the fingerprints. The parameters of this partitioning technique, i.e., the number of clusters and the set of initial cluster centers, are determined in an automated manner by employing the histogram of the fingerprint orientation field. The development of the HA-FCM partitioning scheme is further pursued to devise an enhanced HA-FCM (EAH-FCM) algorithm. In this algorithm, the smoothness of the fingerprint partitioning is improved through a regularization of the fingerprint orientation field, and the computational complexity is reduced by decreasing the number of operations and by increasing the convergence rate of the underlying iterative process of the HA-FCM technique. Finally, a new fuzzy based fingerprint recognition scheme (FMSFR), based on the EHA-FCM partitioning scheme and the basic ideas used in the development of the MSFR scheme, is proposed.udExtensive experiments are conducted throughout this thesis using a number of challenging benchmark databases. These databases are selected from the FVC2002, FVC2004 and FVC2006 competitions containing a wide variety of challenges for fingerprint recognition. Simulation results demonstrate not only the effectiveness of the proposed techniques and schemes but also their superiority over some of the state-of-the-art techniques, in terms of the recognition accuracy and the computational complexity.