Handwritten signature verification employing dynamic time warping

Joanna Maria Putz-Leszczyńska

Abstract

Many handwritten signature verification algorithms have been developed in order to distinguish between genuine signatures and forgeries. An important group of these meth- ods is based on Dynamic Time Warping (DTW). Traditional use of DTW for signature verification consists of forming some misalignment score between the verified signature and a set of template signatures. The right selection of template signatures has a big impact on that verification. In this dissertation, we propose replacing the template signa- tures with the hidden signature - an artificial signature which is created by minimizing the mean misalignment between this signature and the signatures from the training set. The hidden signature opens a number of new possibilities for signature analysis. We proposed and tested several methods for hidden signature estimation. We used statistical properties of the hidden signature to normalize the error signal of verified signature and to use the misalignment on the normalized errors as a verification base. Additionally, we tested this method for different sizes of training sets that are used for the template creation. The second main research direction was also related to the template creation. We propose the universal forgery features idea, where the global classifier is able to classify a signature as a genuine one or, as a forgery, without the actual knowledge of the signature template and its owner. This classifier is learnt once, during the system tuning on a group of historical data. A global classifier trained on a set of training signatures would not be additionally trained after implementation; in other words, additional users enrollments would have no effect on the global classifier parameters. Finally, we combined the idea of hidden signature with the universal forgery idea, thus obtaining a final algorithm. We showed that a global classifier built on the univer- sal forgery feature idea can be also regarded as an user–adjusted threshold. Both ideas allowed us to create a signature verification system with low system requirements that works fast and shows a low level of verification errors. As a result, it can be implemented on mobile platforms and emended devices.
Diploma typeDoctor of Philosophy
Author Joanna Maria Putz-Leszczyńska (FEIT / AK)
Joanna Maria Putz-Leszczyńska,,
- The Institute of Control and Computation Engineering
Title in English Handwritten signature verification employing dynamic time warping
Languageen angielski
Certifying UnitFaculty of Electronics and Information Technology (FEIT)
Disciplineinformation science / (technology domain) / (technological sciences)
Defense Date21-12-2010
End date21-12-2010
Supervisor Andrzej Pacut (FEIT / AK)
Andrzej Pacut,,
- The Institute of Control and Computation Engineering

Internal reviewers Khalid Saeed
Khalid Saeed,,
-

Władysław Skarbek (FEIT / IRMT)
Władysław Skarbek,,
- The Institute of Radioelectronics and Multimedia Technology
Pages134
Keywords in English biometrics, signature verification, dynamic time warping, hidden signature
Abstract in EnglishMany handwritten signature verification algorithms have been developed in order to distinguish between genuine signatures and forgeries. An important group of these meth- ods is based on Dynamic Time Warping (DTW). Traditional use of DTW for signature verification consists of forming some misalignment score between the verified signature and a set of template signatures. The right selection of template signatures has a big impact on that verification. In this dissertation, we propose replacing the template signa- tures with the hidden signature - an artificial signature which is created by minimizing the mean misalignment between this signature and the signatures from the training set. The hidden signature opens a number of new possibilities for signature analysis. We proposed and tested several methods for hidden signature estimation. We used statistical properties of the hidden signature to normalize the error signal of verified signature and to use the misalignment on the normalized errors as a verification base. Additionally, we tested this method for different sizes of training sets that are used for the template creation. The second main research direction was also related to the template creation. We propose the universal forgery features idea, where the global classifier is able to classify a signature as a genuine one or, as a forgery, without the actual knowledge of the signature template and its owner. This classifier is learnt once, during the system tuning on a group of historical data. A global classifier trained on a set of training signatures would not be additionally trained after implementation; in other words, additional users enrollments would have no effect on the global classifier parameters. Finally, we combined the idea of hidden signature with the universal forgery idea, thus obtaining a final algorithm. We showed that a global classifier built on the univer- sal forgery feature idea can be also regarded as an user–adjusted threshold. Both ideas allowed us to create a signature verification system with low system requirements that works fast and shows a low level of verification errors. As a result, it can be implemented on mobile platforms and emended devices.
Thesis file
366-10 doktorat Putz-Leszczynska.pdf 2.45 MB

Get link to the record

Back
Confirmation
Are you sure?