Système d'identification de personnes basé sur la rétine

Abstract : Our work is part of the retina biometrics. The retina is the sensory layer of the eye; it has a rich and unique texture even in twins. Its properties have made the retina biometrics an active research area. Indeed, numerous methods have been proposed for the various stages of the biometric method, from pretreatment of the retinal image to its analysis, through its characterization, in order to identify and authenticate an individual. We are interested in this work in these thesis works, the study, design, development and evaluation of a new biometric method based on the retina. This thesis presents our contributions for each stage of the proposed biometric method. Our first contribution lies in the proposition of a healthy and pathological retinal image analysis method, followed by a selection of a region of interest around the optical disc. This method improves the quality of the retinal image and extracts a more stable region of interest from the retina to maintain a satisfactory information density, to ensure a better quality of recognition. Our second contribution consists in proposing a new method for extracting local characteristics based on the standard SIFT descriptor. It applies a new method based on the removal of non-informative points of interest extracted by the standard SIFT descriptor. This new method of extracting local features reduces the number of redundant points of interest while maintaining the quality of the description. We validated, the proposed biometric method on different bases including healthy and pathological images. This biometric method has yielded encouraging results on healthy and pathological retinal images. The results obtained show encouraging performances. These results indicate that the method we have proposed, correctly locates the retinal region of interest. In identification mode, a correct identification rate of approximately 99.8% is reached. In verification mode, we obtained 0.12% as FRR error rate and 0% for the FAR and EER error rates. The comparative study showed that our method is more discriminative than other state-of-the-art methods, especially those based on segmentation and extraction of the vascular tree
Document type :
Theses
Complete list of metadatas

Cited literature [34 references]  Display  Hide  Download

https://pastel.archives-ouvertes.fr/tel-02085887
Contributor : Abes Star <>
Submitted on : Sunday, March 31, 2019 - 10:20:11 PM
Last modification on : Saturday, May 18, 2019 - 12:27:26 AM

File

TH2018PESC1145.pdf
Version validated by the jury (STAR)

Identifiers

  • HAL Id : tel-02085887, version 1

Collections

Citation

Takwa Chihaoui. Système d'identification de personnes basé sur la rétine. Traitement du signal et de l'image [eess.SP]. Université Paris-Est; Université de Tunis El Manar, 2018. Français. ⟨NNT : 2018PESC1145⟩. ⟨tel-02085887⟩

Share

Metrics

Record views

114

Files downloads

44