Sustav povećanja razlučivosti slike zasnovan na strojnom učenju

master's thesis

• Title (english): Image Super-Resolution System Based on Machine Learning
• Author: Karlo Anđelković
• Mentor: Marin Šilić (mentor)
• Committee member: Marin Šilić (predsjednik povjerenstva)
• Committee member: Goran Delač (član povjerenstva)
• Committee member: Klemo Vladimir (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2024-07-04, Croatia

Abstract

Super rezolucija je postupak dobivanja slike visoke rezolucije iz slike niske rezolucije. Duboko je učenje najpopularniji i najkvalitetniji pristup rješavanju problema te postoje raznolike tehnike kojima se može pristupiti zadatku. U odabiru arhitektura ističu se generativne suparničke mreže i transformeri, a za funkciju gubitka često se uzimaju pikselni, perceptivni i gubitak teksture. Za zadatak je ključno imati dobar i opsežan skup podataka te procijeniti uspješnost modela prikladnom evaluacijskom metrikom poput $PSNR$ ili $SSIM$ mjere. Pri izradi modela važno je odabrati dobru ravnotežu između stvaranja vizualno ugodnih i objektivno sličnih rezultata. Područje super rezolucije već postiže izvrsne rezultate, ali napredak u brojnim aspektima je i dalje moguć.

Abstract (english)

Super resolution is the process of obtaining a high-resolution image from its low-resolution counterpart. Deep learning is the most popular and high-quality approach to solving this problem, and there are various techniques that can be applied to the task. In the selection of architectures, generative adversarial networks and transformers stand out, and for the loss function, pixel, perceptual, and texture loss are often used. For this task, it is crucial to have a good and extensive dataset and to assess the success of the model with an appropriate evaluation metric such as PSNR or SSIM measures. When creating the model, it is important to strike a good balance between generating visually pleasing and objectively similar results. The field of super resolution is already achieving excellent results, but progress in many aspects is still possible.

• Keywords: super rezolucija; duboko učenje; naduzorkovanje; generativni modeli
• Keywords (english): super resolution; deep learning; upsampling; generative models
• Language: croatian
• URN:NBN: urn:nbn:hr:168:390474
• 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: 2024-10-12 19:58:48