magistrsko delo
Aljaž Đukić (Author), Vitomir Štruc (Mentor)

Abstract

Človeška šarenica predstavlja eno od biometričnih modalnosti, ki omogočajo najbolj natančno in zanesljivo razpoznavanje, ter se pogosto uporablja v sistemih dokazovanja istovetnosti. Ključen korak predprocesiranja za doseganje kakovostne in točne razpoznave šarenic je segmentacija, ki določi, na katerem delu zajete slike se šarenica nahaja. Pristopi segmentacije šarenic so se v zadnjih letih premaknili od tradicionalnih algoritmov k metodam globokega učenja, ki imajo pred njimi mnoge prednosti. V magistrskem delu sledimo metodam segmentacije šarenic z uporabo globokega učenja ter z izvedbo pristopa večopravilnega učenja poskušamo še dodatno izboljšati kakovost dosežene segmentacije. V ta namen razvijemo in preverimo delovanje različnih modelov enoopravilnega in večopravilnega učenja, katerih arhitektura temelji na omrežju U-Net, ki ga za potrebe našega dela še dodatno prilagodimo. Ovrednotimo tudi vpliv, ki ga ima na kakovost segmentacije uporaba različnih pomožnih opravil in uteži izgubnih funkcij. Poleg pomožnega opravila dokončanja slik za referenco preverimo še delovanje modelov z uporabo pomožnih opravil odstranjevanja šuma s slik ter barvanja sivinskih slik. Izbrane modele učimo in preverjamo na podatkovnih zbirkah MOBIUS in SBVPI, kjer med pomožnimi opravili modelov večopravilnega učenja dokončanje slik doseže najboljše delovanje na obeh podatkovnih zbirkah. Do izboljšave rezultatov segmentacije enoopravilnega učenja s pristopom večopravilnega učenja s pomožno nalogo dokončanja slik pride le pri preverjanju na podatkovni zbirki SBVPI, kar pripišemo razlikam med samima zbirkama. Prav tako pokažemo, da izbira prevelikih uteži izgubnih funkcij pomožnih opravil vodi do slabše kakovosti segmentacije zaradi povečanega vpliva le-teh na učenje modelov.

Keywords

segmentacija šarenice;večopravilno učenje;dokončanje slik;globoko učenje;U-Net;CNN;biometrija;računalniški vid;magisteriji;

Data

Language: Slovenian
Year of publishing:
Typology: 2.09 - Master's Thesis
Organization: UL FE - Faculty of Electrical Engineering
Publisher: [A. Đukić]
UDC: 004.9:004.8(043.3)
COBISS: 105851139 Link will open in a new window
Views: 135
Downloads: 33
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: Development of a multitask model for iris segmentation
Secondary abstract: The human iris is considered an extremely safe and reliable physiological modality and is thus often used in biometric recognition systems. A crucial pre-processing step for reliable and accurate iris recognition lies in iris segmentation, a process that determines which part of the captured image belongs to the iris. Iris segmentation has in recent years shifted from traditional algorithms to deep learning approaches, which have many advantages. In our work, we follow the trend of using deep learning for solving the task of iris segmentation as we try to further improve the achieved accuracy of iris segmentation using multi-task learning. For this purpose, we develop and evaluate different single-task and multi-task learning models, whose architecture is based on the classic U-Net network, which we additionally modify. We also assess the effect of using different auxiliary tasks and loss weights on the iris segmentation accuracy. Besides the auxiliary task of image inpainting, we also evaluate the performance of models built using the auxiliary tasks of image denoising and colourization of grey images. The chosen models are trained and evaluated on the MOBIUS and SBVPI datasets, where the auxiliary task of image inpainting achieves the best performance among the tested multi-task learning auxiliary tasks on both datasets. The iris segmentation performance of the single-task learning model is improved by using the multi-task learning model with image inpainting chosen as the auxiliary task only when evaluated on the SBVPI dataset, which we contribute to the differences between the datasets. We also demonstrate that choosing bigger auxiliary tasks' loss weights adversely impacts the performance of iris segmentation because of their increased influence on the training of the models.
Secondary keywords: iris segmentation;multi-task learning;image inpainting;deep learning;U-Net;CNN;biometrics;computer vision;
Type (COBISS): Master's thesis/paper
Study programme: 1000316
Embargo end date (OpenAIRE): 1970-01-01
Thesis comment: Univ. v Ljubljani, Fak. za elektrotehniko
Pages: XVIII, 66 str.
ID: 15088703