Non-invertible online signature biometric template protection via shuffling and trigonometry transformation

This thesis reports on research work carried out on online handwritten signature biometrics template protection. Such a research effort is deemed necessary since like any other biometrics system, online signature biometric requires the use of template database which is vulnerable to attacks. The proposed template protection scheme is based on a hybrid of shuffling and trigonometry transformation functions which is addressed as ‘BioTrigono’. Prior to template protection scheme, signature samples are normalized in length to 256 sampled points. The effectiveness of the template protection scheme has been tested via an online signature verification system on SIGMA database which consists of a huge set of signature samples belonging to 200 Malaysian nationals. In the online signature verification system, prominent features are extracted via Principal Component Analysis, which are later fed to the Artificial Neural Network module for signature classification purposes. Here, the final output is accept or reject decision on the status of the tested signature sample as belonging to the claimed reference identity. In this thesis, the noninvertible property of BioTrigono template protection scheme has been proven mathematically through the employment of trigonometry transform function. In addition, renewability testing has been conducted by plotting various Receiver Operating Characteristic (ROC) curves under different settings which results that BioTrigono template protection scheme has renewability property. The impact of BioTrigono template protection scheme on the accuracy of the online signature verification system has also been conducted and benchmarked against an existing template protection mechanism (i.e. BioConvolving). The verification results indicate that the online signature verification system has consistently produced a better system accuracy when used together with BioTrigono template protection scheme as compared to its benchmarked counterpart.

Preview PDF FK 2014 2R.pdf Download (1MB) | Preview

Actions (login required)

View Item