Navidezno pomerjanje frizur s postopki računalniškega vida

Lipovšek, Jan (author), Štruc, Vitomir (mentor)

V delu je predstavljena aplikacija za navidezno pomerjanje frizur s postopki računalniškega vida. Delo predstavlja problem navideznega pomerjanja frizur z uporabo dveh različnih metod segmentacije, ki vhodno sliko osebe razdelita v tri semantična področja (lase, obraz in ozadje), ter dveh metod vrisovanja, ki na sliki po zamenjavi frizur smiselno zapolnita luknje. Na podatkovni zbirki LFW smo naučili in ovrednotili dve globoki segmenta- cijski nevronski mreži FCN-8s in U-Net. Rezultati kažejo, da je za namen se- gmentacije las bolj primerna mreža U-Net, z natančnostjo po slikovnih elementih 0.910 v primerjavi z natančnostjo 0.902, ki jo je dosegla mreža FCN-8s. Izračun F-mere prav tako potrdi zgornjo trditev. Za vrisovanje smo preizkusili globoko generativno nevronsko omrežje in postopek Telea, ki temelji na polnjenju lukenj glede na vrednosti okoliških slikovnih pik. S povprečno oceno 4.214 pri testih s človeškimi ocenjevalci, v primerjavi s 3.429, se je za namen vrisovanja bolje izkazala metoda Telea. Poglavje 1 predstavi uvod v delo. Poglavje 2 predstavi teoretično ozadje za delom, ter pregled podobnih aplikacij. V poglavju 3 so opisane uporabljene metode. Poglavje 4 poda rezultate segmentacije, normalizacije, združevanja slik in vrisovanja, ter ovrednotenje zgornjih rezultatov. Zaključki pa so podani v poglavju 5.

• Language: Slovenian
• Keywords: globoko učenje, računalniški vid, konvolucijske nevronske mreže, navidezno pomerjanje frizur
• Work type: Bachelor thesis/paper
• Organization: FE - Faculty of Electrical Engineering
• Year: 2018
• PID: 20.500.12556/RUL-102745
• Publication date in RUL: 07.09.2018

LIPOVŠEK, Jan, 2018, Navidezno pomerjanje frizur s postopki računalniškega vida [online]. Bachelor’s thesis. [Accessed 30 March 2025]. Retrieved from: https://repozitorij.uni-lj.si/IzpisGradiva.php?lang=eng&id=102745

Secondary language

• Language: English
• Title: Virtual hair makeover using computer vision tools

Abstract

The thesis presents an application designed to virtually fit different hairstyles to an image using computer vision tools. It uses two methods of segmentation, which divide the input image into three separate semantic regions (hair, face and background) and two methods of inpainting, which fill the holes left in the process of combining hair with faces. The neural networks FCN-8s and U-Net used in this work, are trained and validated on the LFW dataset. The results indicate, that U-Net outperforms FCN-8s in hair segmentation tasks, where it achieved a higher average per-pixel accuracy of 0.910, where FCN-8s’ average per-pixel accuracy was 0.902, respectively. The F-score results, where U-Net achieved better accuracy across all classes, further confirms the previous observations. We used a deep generative neural network and the Thelea method, which fills holes according to its neighboring pixels, for inpainting. The Thelea method achieved a slightly better average score of 4.214 in the human rating test, as opposed to 3.429 of the deep generative neural network. Chapter 1 provides an introduction to the thesis. The first part of Chapter 2 presents a short overview of the fundamental elements in deep learning, whereas the second part explores related work. Chapter 3 explains the methodology behind the application. Chapter 4 presents the results, and validates the segmentation and inpainting methods. Chapter 5 presents the conclusion and further work.

• Keywords: deep learning, computer vision, convolutional neural networks, hair