Biométrie multimodale basée sur l’iris et le visage

Abstract : This thesis aims to make a contribution in the field of biometrics based on iris (one of the most accurate and hard to hack biometrics) in conjunction with face (one of the cheapest and less intrusive biometrics).Through this work, we discuss several important aspects of unimodal and multimodal biometrics. After an overview on unimodal and multimodal biometrics based on iris and face, we propose several personal approaches of biometric authentication using each single trait. Particularly, we address facial recognition first with conventional approaches based on combined algorithms, then with bio-inspired approaches emulating the human vision mechanism. We demonstrate the interest of bio-inspired approaches over conventional approaches through two proposed methods. The first one exploits the results of neuroscientific work indicating the relevant regions and scales in a face identification task. The second consists in applying a rank order coding method at the preprocessing step so as to enhance the information content of face images.We keep the best unimodal approach of iris and face recognition to design two multimodal biometric methods. Through these methods, we evaluate different classic strategies of multimodal score-level fusion. Afterwards, we propose a new score-level fusion rule based on a quality metric according to irises occultation rates. Then, we point out the interest of the double-sample iris aspect in a multimodal approach.All the proposed methods are evaluated on the real multimodal IV² database captured under variable to degraded environments, and following a specific protocol provided as part of the IV² evaluation campaign. After a comparative study with the participant algorithms in the IV² campaign, we prove the competitiveness of our algorithms witch outperform most of the participant ones in the IV² campaign in many experiments.
Complete list of metadatas

Cited literature [217 references]  Display  Hide  Download

https://hal.archives-ouvertes.fr/tel-01762973
Contributor : Frédéric Davesne <>
Submitted on : Tuesday, April 10, 2018 - 3:31:48 PM
Last modification on : Monday, October 28, 2019 - 10:50:22 AM

File

Manuscrit_de_these_NKH.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : tel-01762973, version 1

Citation

Nefissa Khiari Hili. Biométrie multimodale basée sur l’iris et le visage. Traitement du signal et de l'image [eess.SP]. Université Paris-Saclay; Université de Tunis El Manar, 2016. Français. ⟨NNT : 2016SACLE014⟩. ⟨tel-01762973⟩

Share

Metrics

Record views

113

Files downloads

1261