Contactless 3D fingerprint identification and contactless to contact-based fingerprint matching

Abstract

Fingerprint, as one of the most accurate and discriminative biometrics, has been investigated and used to identify human beings for hundreds of years. Fingerprint identifcation systems have also been developed and widely deployed for forensics, security and civilians applications for more than decades. In order to address the limitations of existing contact-based fingerprint identifcation systems and improve recognition accuracy, contactless 3D fingerprint recognition technologies have attracted growing attention. However, the complex 3D imaging setups employed in these systems typically require structured lighting with scanners or multiple cameras which are bulky and with higher cost. In addition, the high time complexity of 3D fingerprint alignment and matching algorithm has become the obstacle for the system deployment. In this Thesis, a more accurate and effcient 3D fingerprint identifcation approach is developed using a single 2D camera with multiple colored light-emitting diode (LED) illumination to address the limitations of existing contactless 3D fingerprint acquisition systems. A 3D minutiae tetrahedron based algorithm is developed to more effciently match recovered minutiae features in 3D space and address the limitations of 3D minutiae matching approach in the literature. This algorithm signifcantly improves the matching time to about 15 times than state-of-the-art in the reference. A hierarchical tetrahedron matching scheme is also developed to further improve the matching accuracy with faster speed. The 2D images acquired to reconstruct the 3D fingerprints are also used to recover 2D minutiae and further improve matching performance for 3D fingerprints. A new two-session database acquiring from 300 different clients consists of 2760 3D fingerprints reconstructed from 5520 colored 2D fingerprints is also developed and shared in public domain to further advance much needed research in this area. Extensive experimental results validate the proposed approach and demonstrate the effectiveness of proposed algorithms. Convolutional neural networks (CNN) have shown remarkable capabilities in biometrics recognition problem. An end-to-end contactless 3D fingerprint representation learning model based on CNN has also been developed to accurately match multi-view contactless 3D fingerprints. The proposed model includes one fully convolutional network for fingerprint segmentation and three Siamese networks for multi-view 3D fingerprints feature representation. Contactless 3D imaging often results in partial 3D fingerprints as it requires relatively higher cooperation from users during the contactless 3D imaging. Such contactless 3D fingerprint images signifcantly degrades matching accuracy due to partial 3D fingerprint imaging. In this Thesis, contactless partial 3D fingerprint identifcation, which is a more challenging problem due to its high degree of freedom during contactless 3D fingerprint acquisition, is also addressed by using proposed model. Such approaches are evaluated on two publicly available databases and state-of-the-art minutiae-based 3D fingerprint recognition methods are used for comparison. Comparative experimental results, presented in this Thesis using state-of-the-art 3D fingerprint recognition method, demonstrate the effectiveness of the proposed multi-view approach and illustrate a signifcant improvement of state-of-the-art 3D fingerprint recognition methods. Accurate matching of contactless 2D fingerprint images with contact-based fingerprints is also critical for the success of emerging contactless 3D/2D fingerprint technologies, which offer more hygienic and deformation-free acquisition of fingerprint features. Inspired by the success of deep learning technologies in images recognition and feature representation, we develop a CNN-based framework to accurately match contactless and contact-based fingerprint images. Our framework firstly trains a multi-Siamese CNN using fingerprint minutiae, respective ridge map and specifc region of ridge map. This network is used to generate deep fingerprint representation using a new loss function. Deep fingerprint representations generated in such multi-Siamese network are concatenated for more accurate cross matching. The proposed approach for cross-fingerprint matching is evaluated on two public databases containing contactless 2D fingerprints and respective contact-based fingerprints. Our experiments consistently achieve outperforming results, over several popular deep learning architectures and over contactless to contact-based fingerprints matching methods in the literature.