Please use this identifier to cite or link to this item: http://theses.ncl.ac.uk/jspui/handle/10443/3982
Full metadata record
DC FieldValueLanguage
dc.contributor.authorAl-Nima, Raid Rafi Omar-
dc.date.accessioned2018-09-10T14:25:04Z-
dc.date.available2018-09-10T14:25:04Z-
dc.date.issued2017-
dc.identifier.urihttp://hdl.handle.net/10443/3982-
dc.descriptionPhD Thesisen_US
dc.description.abstractIn recent years, Finger Textures (FTs) have attracted considerable attention as potential biometric characteristics. They can provide robust recognition performance as they have various human-speci c features, such as wrinkles and apparent lines distributed along the inner surface of all ngers. The main topic of this thesis is verifying people according to their unique FT patterns by exploiting signal processing and machine learning techniques. A Robust Finger Segmentation (RFS) method is rst proposed to isolate nger images from a hand area. It is able to detect the ngers as objects from a hand image. An e cient adaptive nger segmentation method is also suggested to address the problem of alignment variations in the hand image called the Adaptive and Robust Finger Segmentation (ARFS) method. A new Multi-scale Sobel Angles Local Binary Pattern (MSALBP) feature extraction method is proposed which combines the Sobel direction angles with the Multi-Scale Local Binary Pattern (MSLBP). Moreover, an enhanced method called the Enhanced Local Line Binary Pattern (ELLBP) is designed to e ciently analyse the FT patterns. As a result, a powerful human veri cation scheme based on nger Feature Level Fusion with a Probabilistic Neural Network (FLFPNN) is proposed. A multi-object fusion method, termed the Finger Contribution Fusion Neural Network (FCFNN), combines the contribution scores of the nger objects. The veri cation performances are examined in the case of missing FT areas. Consequently, to overcome nger regions which are poorly imaged a method is suggested to salvage missing FT elements by exploiting the information embedded within the trained Probabilistic Neural Network (PNN). Finally, a novel method to produce a Receiver Operating Characteristic (ROC) curve from a PNN is suggested. Furthermore, additional development to this method is applied to generate the ROC graph from the FCFNN. Three databases are employed for evaluation: The Hong Kong Polytechnic University Contact-free 3D/2D (PolyU3D2D), Indian Institute of Technology (IIT) Delhi and Spectral 460nm (S460) from the CASIA Multi-Spectral (CASIAMS) databases. Comparative simulation studies con rm the e ciency of the proposed methods for human veri cation. The main advantage of both segmentation approaches, the RFS and ARFS, is that they can collect all the FT features. The best results have been benchmarked for the ELLBP feature extraction with the FCFNN, where the best Equal Error Rate (EER) values for the three databases PolyU3D2D, IIT Delhi and CASIAMS (S460) have been achieved 0.11%, 1.35% and 0%, respectively. The proposed salvage approach for the missing feature elements has the capability to enhance the veri cation performance for the FLFPNN. Moreover, ROC graphs have been successively established from the PNN and FCFNN.en_US
dc.description.sponsorshipthe ministry of higher education and scientific research in Iraq (MOHESR); the Technical college of Mosul; the Iraqi Cultural Attach e; the active people in the MOHESR, who strongly supported Iraqi students;en_US
dc.language.isoenen_US
dc.publisherNewcastle Universityen_US
dc.titleSignal processing and machine learning techniques for human verification based on finger texturesen_US
dc.typeThesisen_US
Appears in Collections:School of Electrical and Electronic Engineering

Files in This Item:
File Description SizeFormat 
Al-Nima, R.R.O. 2017.pdfThesis19.3 MBAdobe PDFView/Open
dspacelicence.pdfLicence43.82 kBAdobe PDFView/Open


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.