Povećavanje rezolucije slika ljudskih lica korištenjem neuronskih mreža

master's thesis

• Title (english): Enhancing the resolution of human face images using neural networks
• Author: Jan Grgić
• Mentor: Igor Mekterović (mentor)
• Committee member: Igor Mekterović (predsjednik povjerenstva)
• Committee member: Ljiljana Brkić (član povjerenstva)
• Committee member: Slaven Zakošek (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2023-07-06, Croatia

Abstract

U radu je razvijeno nekoliko dubokih konvolucijskih i generativnih suparničkih mreža. Prilikom treniranja, isprobavane su različite funkcije gubitka kako bi se istražio njihov utjecaj na performanse modela. Metrike PSNR i SSIM su korištene kao kvantitativne mjere za ocjenu kvalitete generiranih slika. Kroz eksperimente, utvrđeno je da funkcije gubitka igraju ključnu ulogu u postizanju visoke kvalitete generiranih slika. Usporedbom različitih funkcija gubitka, uočene su razlike u iznosu metrika, kao i u izgledu generiranih slika. Posebnu pozornost privukla je mreža GAN-VGG22 koja koristi izlaz sloja block3_conv4 te postiže dobre rezultate. Na kraju je razvijena web aplikacija koja koristi mrežu GAN-VGG22 za povećanje razlučivosti slika.

Abstract (english)

This paper presents the development of several deep convolutional and generative adversarial networks. The impact of different loss functions on model performance was investigated during the training process, with PSNR and SSIM metrics used for quantitative evaluation of the generated images quality. Experimental results revealed the crucial role of loss functions in achieving high-quality generated images, as evidenced by variations in metric values and the visual appearance of the generated images. Special attention was attracted by the GAN-VGG22 network, which uses the output of the block3_conv4 layer and achieves good results. Finally, a web application was developed that uses the GAN-VGG22 network to increase the resolution of images.

• Keywords: Povećanje razlučivosti; duboke konvolucijske neuronske mreže; generativne suparničke mreže; superrazlučivost; ljudsko lice; PSNR; SSIM; funkcije gubitka
• Keywords (english): Image upscaling; deep convolutional neural networks; generative adversial networks; superresolution; human face; PSNR; SSIM; loss functions
• Language: croatian
• URN:NBN: urn:nbn:hr:168:442540
• Study programme: Title: Computing; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka računarstva (magistar/magistra inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2023-10-11 08:57:20