Segmentacija očesnega ožilja iz slikovnih podatkov z globokim konvolucijskim omrežjem

ROT, ŽIGA (author), Štruc, Vitomir (mentor)

V tem delu predstavimo izvedbo mehanizma, ki iz slik očesa izloča ožilje beločnice – biometrično karakteristiko, ki se lahko uporabi v sistemih za razpoznavanje šarenice za izboljšanje zanesljivosti in natančnosti razpoznavanja. Model sestavljata dve stopnji. Prva skrbi za izločanje območja zanimanja – beločnice, iz katere nato druga stopnja segmentira ožilje. Našo izvedbo utemeljimo z dvema eksperimentoma. V prvem pokažemo vpliv izločanja območja zanimanja na končni izhod. V drugem prikažemo razliko v uspešnosti segmentacije med enorazredno in večrazredno segmentacijo beločnice.

• Language: Slovenian
• Keywords: segmentacija slik, ožilje beločnice, globoke konvolucijske nevronske mreže, U-net
• Work type: Bachelor thesis/paper
• Organization: FE - Faculty of Electrical Engineering
• Year: 2022
• PID: 20.500.12556/RUL-139570
• COBISS.SI-ID: 120224771
• Publication date in RUL: 05.09.2022

Secondary language

• Language: English
• Title: Segmentation of ocular vasculature from visual data with a deep convolutional network

Abstract

This thesis presents the implementation of a model that can extract scleral vasculature from image data – a biometric feature which can be used in iris-based biometric recognition systems to enhance robustness and accuracy. The model consists of two stages. The first stage is used for extraction of the region of interest – the sclera, from which then the next stage segments the vascular structure. We justify our design with two experiments. In the first one, we show the impact of prior extraction of the region of interest on the final output. In the second one, we present the difference in segmentation quality between binary and multi-class versions of sclera segmentation.

• Keywords: image segmentation, sclera vascularity, deep convolutional neural networks, U-net