Duboki modeli za uklanjanje šuma u slikama iscrtanim slučajnim uzorkovanjem

master's thesis

• Title (english): Deep models for denoising images obtained by Monte Carlo sampling
• Author: Nikola Bunjevac
• Mentor: Siniša Šegvić (mentor)
• Committee member: Siniša Šegvić (predsjednik povjerenstva)
• Committee member: Željka Mihajlović (član povjerenstva)
• Committee member: Marko Čupić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2019-07-15, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Iscrtavanje slučajnim uzorkovanjem postiglo je zavidne rezultate od svojih začetaka. Napretkom tehnologije omogućena su dodatna poboljšanja kvalitete generiranih slika. Ipak,
značajan dio vremena troši se na veliki broj uzoraka koji je nužan za željenu kvalitetu rezultata. Nedovoljan broj uzoraka manifestira se u obliku šuma. Ovaj rad razmatra alternativni način uklanjanja šuma primjenom dubokog učenja. Spomenuti pristup pogodan je zbog toga što možemo generirati parove slika s i bez prisutnog šuma. Korišteni su duboki konvolucijski modeli, a kao dodatno poboljšanje korišten je pristup uklanjanja šuma pomoću jezgre koji se pokazao efikasnijim od naivnog pristupa. Prezentirani su postignuti rezultati te predloženi pravci budućeg razvoja.

Abstract (english)

Monte Carlo rendering has made remarkable results since its inception. Technological
progress has enabled additional improvements to the quality of the generated images. Nevertheless, a significant part of computing time is spent on large number of samples that are necessary for the desired quality of results. Insufficient number of samples is manifested
in the form of a noise. This thesis deals with an alternative way of denoising by applying
deep learning. This approach is appropriate because we can generate pairs of pictures with
and without noise. Deep convolutional models were used, and in addition a kernel based
approach was used for noise removal that proved to be more efficient than a naive approach.
Results and the proposed directions for future development are presented.

• Keywords: Duboko učenje; računalna grafika; algoritam praćenja zrake; slučajno uzorkovanje
• Keywords (english): Deep learning; computer graphics; ray tracing; Monte Carlo method
• Language: croatian
• URN:NBN: urn:nbn:hr:168:200138
• 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: 2020-01-07 17:09:15