Prenos stila slike s pomočjo prenosnega učenja in nevronskih mrež

magistrsko delo

Ivona Čolaković (Author), Sašo Karakatič (Mentor)

Abstract

Hitro razvijajoče področje umetne inteligence se v zadnjih letih integrira v različna področja in tako postaja neizogiben del številnih človeških dejavnosti. Umetna inteligenca je pokazala, da se lahko integrira tudi v področje umetnosti in ustvarja nova umetniška dela a podlagi kopiranja stilov grafičnih del priznanih avtorjev. Nevronske mreže, ki posnamejo delovanje človeških možganov, dodatno pomagajo pri tem postopku, saj omogočajo razpoznavo vzorcev v stilih grafičnih del. V magistrskem delu se osredotočimo na raziskovanje tehnike prenosa stila grafičnih del iz enega na drugo grafično delo s pomočjo nevornskih mrež. V ta namen opišemo sestavne dele nevronskih mrež, podrobneje razložimo konvolucijske nevronske mreže in predstavimo pojem prenosnega učenja. Z namenom boljšeg razumevanja področja prenosa stila ilustracij pregledamo obstoječe raziskave ter opišemo delovanje algoritma za prenos stila. V okviru magistrskega dela prikažemo implementacijo in rezultate eksperimenta skozi katerega smo ugotovili, da pristop prenosa stila lahko uspešno prenaša stil iz ilustracij na fotografije kakor tudi iz ilustracij na druge ilustracije.

Keywords

prenos stila;konvolucijske nevronske mreže;prenosno učenje;umetna inteligenca;umetnost;razpoznavanje slik;magistrske naloge;

Data

Language:	Slovenian
Year of publishing:	2021
Typology:	2.09 - Master's Thesis
Organization:	UM FERI - Faculty of Electrical Engineering and Computer Science
Publisher:	[I. Čolaković]
UDC:	659.113.7:007.774.1(043.2)
COBISS:	82420227
Views:	293
Downloads:	56
Average score:	0 (0 votes)
Metadata:

Other data

Secondary language:	English
Secondary title:	Image style transfer using transfer learning and neural networks
Secondary abstract:	Artificial Intelligence, a rapidly evolving field, has became an inevitable part of many human activities due to integration into different field. Artificial Intelligence has shown it can be integrated even into art by creating new works of art based on copying works of renowned authors. Neural networks which imitate the functioning of human brain further help in this field as they are able to recognize patterns in images. In Master’s thesis we focus on researching image style transfer techniques. For this purpose we describe neural networks components, explain convolutional neural networks and introduce the concept of transfer learning. In order to better understand the field of style transfer, we review existing research and describe the style transfer algorithm. In Master’s thesis we show the implementation and results of experiment that helped us conclude that approach of style transfer can be used to successfully transfer style from image to the photography or ilustrations.
Secondary keywords:	style transfer;convolutional neural networks;transfer learning;
Type (COBISS):	Master's thesis/paper
Thesis comment:	Univ. v Mariboru, Fak. za elektrotehniko, računalništvo in informatiko, Informatika in tehnologije komuniciranja
Pages:	XI, 56 str.
ID:	13280147