doktorska disertacija
Marko Kočevar (Author), Zdravko Kačič (Mentor)

Abstract

Izboljšanje slike prstnega odtisa je ključni korak v sistemu samodejnega prepoznavanja prstnih odtisov (angl. Automatican Fingerprint System - AFIS). Zaradi slabe kvalitete prstnega odtisa lahko algoritem izločanja značilk nepravilno izloči značilke, ki nato vplivajo na nepravilno ujemanje prstnih odtisov in posledično na neučinkovito prepoznavo oseb na podlagi prstnih odtisov. Tehnike izboljšanja slike prstnega odtisa temeljijo na izboljšanju v prostorski ter frekvenčni domeni ali pa kombinaciji obeh. V doktorski disertaciji je definiran postopek dvostopenjskega adaptivnega izboljšanja slike prstnega odtisa nizke kvalitete s kontekstnimi filtri, uporabljenimi na prvi stopnji (v prostorski domeni) in dodatno na drugi stopnji (v frekvenčni domeni). Kontekstne filtre uvrščamo med usmerjene filtre, ki so krmiljeni na osnovi kontekstnih informacij (frekvenca in orientacija lokalnega grebena) v sliki prstnega odtisa. Ti filtri lahko spreminjajo strukturo grebena in dolin, kar pomeni, da lahko povežejo prekinjene grebene oz. ločijo povezane grebene. Vendar pa je potrebno pred izboljšanjem slike prstnega odtisa s kontekstnimi filtri najprej izboljšati kontrast in čitljivost slike z neusmerjenimi filtri. V ta namen smo definirali blokovno lokalno normalizacijo, ki jo optimiziramo adaptivno glede na spremembo intenzitete sivine v originalni sliki prstnega odtisa. Za potrebe kontekstnega izboljšanja na prvi stopnji smo predstavili dve tehniki izboljšanja; usmerjeno linearna anizotropično difuzijo in usmerjeno kompenzacijo lokalnega grebena, ki smo jo prav tako optimizirali z adaptivnim spreminjanjem velikosti maske glede na stopnjo intenzitete sivine v originalni sliki prstnega odtisa. Za potrebe krmiljenja usmerjenega difuzijskega filtra smo orientacijo grebena na osnovi gradienta dodatno izboljšali in definirali usmerjeno polje. Na drugi stopnji sliko prstnega odtisa dodatno izboljšamo s kratkočasovno Fourierjevo transformacijo. Dvostopenjski adaptivni algoritem učinkovito izboljša sliko prstnega odtisa v primerjavi z najsodobnejšimi algoritmi. Eksperimentalni rezultati, izvedeni na javno dostopni bazi prstnih odtisov FVC, so potrdili uspešnost definiranega postopka.

Keywords

slike prstnih odtisov;frekvenčna domena;obdelava slik;izboljšanje slik;prepoznavanje prstnih odtisov;kontekstne informacije;

Data

Language: Slovenian
Year of publishing:
Typology: 2.08 - Doctoral Dissertation
Organization: UM FERI - Faculty of Electrical Engineering and Computer Science
Publisher: [M. Kočevar]
UDC: 004.352.242:343.982(043.3)
COBISS: 278199808 Link will open in a new window
Views: 1626
Downloads: 152
Average score: 0 (0 votes)
Metadata: JSON JSON-RDF JSON-LD TURTLE N-TRIPLES XML RDFA MICRODATA DC-XML DC-RDF RDF

Other data

Secondary language: English
Secondary title: Two-stage adaptive enhancement of a low-quality fingerprint image on the basis of contextual information
Secondary abstract: Fingerprint enhancement is a key step in the Automated Fingerprint Identification System (AFIS). Because of poor quality of fingerprint image, the algorithm for feature extraction may extract features incorrectly, which may lead to incorrect fingerprint matching and consequently to inefficient fingerprint-based identity verification. Techniques of fingerprint image enhancement are based on improvement in the spatial domain, frequency domain, or a combination of both. This thesis presents a two-stage adaptive enhancement of a low quality fingerprint image with contextual filters; initially in the first stage (spatial domain) and additionally in the second stage (frequency domain). Contextual filters are directional filters which are controlled by contextual information of the fingerprint image (local ridge frequency and orientation). These filters can change ridge structures by connecting disconnected ridges and separating connected ridges. However, before improving the fingerprint image with contextual filters, it is necessary to first improve contrast and readability of the fingerprint image with non-directional filters. For this purpose block-local normalization was used, which was optimized adaptively according to original image intensity information. Two methods of enhancement are presented to enhance context information in the first stage: oriented linear anisotropic diffusion and oriented local compensation filter, where the size of the mask is optimized adaptively according to the degree of intensity in the original fingerprint image. For the purposes of control of the oriented diffusion filter, ridge orientation was additionally improved on the basis of the gradient, so that we got a defined orientation field. In the second stage the fingerprint image was additionally enhanced with short-time Fourier transform. Adaptive two-stage algorithm effectively enhances the fingerprint image in comparison to some other state-of-the-art algorithms. Experimental results were obtained on a public fingerprint database FVC.
Secondary keywords: fingerprint enchancement;two-stage;frequency domain;spatial doamin;image processing;fingerprint recognition;Prstni odtisi;Disertacije;Prepoznavanje ujemanja;
URN: URN:SI:UM:
Type (COBISS): Doctoral dissertation
Thesis comment: Univ. v Mariboru, Fak. za elektrotehniko, računalništvo in informatiko
Pages: IX f., 105 str.
ID: 8702791