Abstract/Details

Jedna klasa biometrijskog kriptosistema zasnovanog na konvolucionim neuronskim mrežama

Barzut, Srđan.   Singidunum University (Serbia) ProQuest Dissertations & Theses,  2022. 30946669.

Abstract (summary)

U ovoj doktorskoj disertaciji predložen je novi biometrijski kriptosistem otisaka prstiju baziran na sistemu fazi povezivanja i dubokih konvolucionih neuronskih mreža. Centralni doprinos rada predstavlja novi pristup automatskom izdvajanju obeležja fiksne dužine iz otisaka prstiju, u potpunosti zasnovanom na konvolucionim neuronskim mrežama. Predloženom kvantizacijom obeležja kodovanjem sa dva bita, biometrijski šabloni su prevedeni u binarni domen, što je omogućilo primenu XOR biometrije i razvoj biometrijskog kriptosistema koji se može koristiti za upravljanje ključevima (engl. key-release) ili za zaštitu šablona. Problem varijabilnosti biometrijskih podataka marginalizovan je primenom BCH koda za korekciju grešaka, koji radi na nivou bloka što ga čini otpornim na poznate statističke napade. Predloženi biometrijski kriptosistem sistem može upravljati dužinom ključeva od 265 bita, što zadovoljava potrebe savremenih kriptografskih sistema, uz prihvatljivu marginu EER greške od 1%. Evaluacija eksperimentalnih rezultata potvrđuje značajan napredak u odnosu na druge biometrijske kriptosisteme i sisteme za poređenje otisaka na osnovu njihove teksture.

Alternate abstract:

You are viewing a machine translation of selected content from our databases. This functionality is provided solely for your convenience and is in no way intended to replace human translation. Show full disclaimer

In this doctoral dissertation, a new biometric cryptosystem of fingerprints based on the fuzzy connection system and deep convolutional neural networks is proposed. The central contribution of the work is a new approach to the automatic extraction of fixed-length features from fingerprints, entirely based on convolutional neural networks. By the proposed quantization of the feature by coding with two bits, the biometric templates were translated into the binary domain, which enabled the application of XOR biometrics and the development of a biometric cryptosystem that can be used for key-release or template protection. The problem of biometric data variability is marginalized by the implementation of the BCH error correction code, which operates at the block level making it resistant to known statistical attacks. The proposed biometric cryptosystem system can manage a key length of 265 bits, which meets the needs of modern cryptographic systems, with an acceptable EER error margin of 1%. The evaluation of the experimental results confirms a significant improvement compared to other biometric cryptosystems and systems for comparing fingerprints based on their texture.

Indexing (details)


Business indexing term
Subject
Computer science;
Artificial intelligence
Classification
0984: Computer science
0800: Artificial intelligence
Identifier / keyword
Convolutional neural networks; Biometric cryptosystem; Fuzzy connection system; Fingerprints
URL
https://nardus.mpn.gov.rs/handle/123456789/18971
Title
Jedna klasa biometrijskog kriptosistema zasnovanog na konvolucionim neuronskim mrežama
Alternate title
One Class of Biometric Cryptosystem Based on Convolutional Neural Networks
Author
Barzut, Srđan
Number of pages
6
Publication year
2022
Degree date
2022
School code
9548
Source
DAI-B 85/10(E), Dissertation Abstracts International
ISBN
9798382094311
Advisor
Milosavljević, Milan
Committee member
Adamović, Saša; Kovačević, Branko
University/institution
Singidunum University (Serbia)
University location
Serbia
Degree
Ph.D.
Source type
Dissertation or Thesis
Language
Serbian
Document type
Dissertation/Thesis
Dissertation/thesis number
30946669
ProQuest document ID
3053934879
Copyright
Database copyright ProQuest LLC; ProQuest does not claim copyright in the individual underlying works.
Document URL
https://www.proquest.com/docview/3053934879/abstract/