# Knowledge base: Warsaw University of Technology

Back

## Template protection mechanisms for iris binary codes

### Marcin Chochowski

#### Abstract

This thesis addresses the problem of applying the Renewable Biometric References framework to binary iris codes. Nowadays, where more and more of our assets are virtual, thus a reliable, “personal” authentication is a must. The biometrics is to be the authentication method that will guarantee the security of our assets. This is true as long as it will be implemented in a proper way. Building an authentication system based on biometrics, one needs to take into account not only security threats common to all information systems, but also those specific to biometrics. One of such threats is the security and revocability of biometric references. A general concept of how to do this, called Renewable Biometric References (RBR), was proposed as an ISO standard “Biometric template protection" [49]. This standard, however, does not specify an exact way of implementation. This dissertation is devoted to the problem of application of the Renewable Biometric References framework in iris biometrics, where the biometric templates and biometric query codes have a form of binary, fixed-length vectors. The main challenge in an implementation of highperformance RBR mechanism is the reduction of inner class variability. The more this variability is limited the better RBR mechanism we can obtain. In this thesis, we identified the main sources of inner class variability and proposed methods to reduce it. In particular, we show that the quality of localization is the major source of errors the RBR framework has to cope with. Even changes as small as 1 pixel in localization parameters cause significant changes within the resulting iris code. We also address the problem of reference selection, proving that the majority reference outperforms any other used method. Eventually, we look into the binary iris code in detail showing the strong, inner structure of dependencies. Basing on this, we propose how to optimize the binary iris codes for RBR mechanism, either by masking some of the worst code elements or by introducing a new coding approach. All results were obtained sing the most popular v and successful iris coding algorithm, yet we show that our findings are general and hold for many other coding methodologies. The RBR mechanism is evaluated after sequential application of the proposed modifications, showing that each one leads to the performance improvement.
Record ID
WUT4124048e295a4a81940876fd9dec2661
Diploma type
Doctor of Philosophy
Author
Marcin Chochowski Marcin Chochowski,, The Institute of Control and Computation Engineering (FEIT/AK)Faculty of Electronics and Information Technology (FEIT)
Title in Polish
Template protection mechanisms for iris binary codes
Title in English
Template protection mechanisms for iris binary codes
Language
(en) English
Certifying Unit
Faculty of Electronics and Information Technology (FEIT)
Discipline
automation and robotics / (technology domain) / (technological sciences)
Status
Finished
Defense Date
08-12-2015
Title date
15-12-2015
Supervisor
External reviewers
Krzysztof Ślot Krzysztof Ślot,, External affiliation of publication: Politechnika Łódzka
Khalid Saeed Khalid Saeed,, External affiliation of publication: Bialystok University of Technology
Honored
yes
Pages
120
Keywords in English
iris biometrics, binary iris codes, template protection, renewable biometric references
Abstract in English
This thesis addresses the problem of applying the Renewable Biometric References framework to binary iris codes. Nowadays, where more and more of our assets are virtual, thus a reliable, “personal” authentication is a must. The biometrics is to be the authentication method that will guarantee the security of our assets. This is true as long as it will be implemented in a proper way. Building an authentication system based on biometrics, one needs to take into account not only security threats common to all information systems, but also those specific to biometrics. One of such threats is the security and revocability of biometric references. A general concept of how to do this, called Renewable Biometric References (RBR), was proposed as an ISO standard “Biometric template protection" [49]. This standard, however, does not specify an exact way of implementation. This dissertation is devoted to the problem of application of the Renewable Biometric References framework in iris biometrics, where the biometric templates and biometric query codes have a form of binary, fixed-length vectors. The main challenge in an implementation of highperformance RBR mechanism is the reduction of inner class variability. The more this variability is limited the better RBR mechanism we can obtain. In this thesis, we identified the main sources of inner class variability and proposed methods to reduce it. In particular, we show that the quality of localization is the major source of errors the RBR framework has to cope with. Even changes as small as 1 pixel in localization parameters cause significant changes within the resulting iris code. We also address the problem of reference selection, proving that the majority reference outperforms any other used method. Eventually, we look into the binary iris code in detail showing the strong, inner structure of dependencies. Basing on this, we propose how to optimize the binary iris codes for RBR mechanism, either by masking some of the worst code elements or by introducing a new coding approach. All results were obtained sing the most popular v and successful iris coding algorithm, yet we show that our findings are general and hold for many other coding methodologies. The RBR mechanism is evaluated after sequential application of the proposed modifications, showing that each one leads to the performance improvement.
PKT classification
iris biometrics, binary iris codes, template protection, renewable biometric references
Thesis file

Uniform Resource Identifier
https://repo.pw.edu.pl/info/phd/WUT4124048e295a4a81940876fd9dec2661/
URN
urn:pw-repo:WUT4124048e295a4a81940876fd9dec2661

Confirmation
Are you sure?