Texture Upsampling and Enhancement Using Neural Synthesis

Blom, Roemer (2023) Texture Upsampling and Enhancement Using Neural Synthesis. Bachelor's Thesis, Computing Science.

Preview	Text bCS_2023_BlomRHW.pdf Download (26MB) | Preview
	Text toestemming.pdf Restricted to Registered users only Download (149kB)

Abstract

Visual texture of surfaces plays an important role in human visual perception, particularly in how we discern and interpret material properties and spatial relationships in our environment. Texture synthesis is the process of algorithmically constructing a novel image texture from a sample, maintaining the visual appearance and essential characteristics of the sample. Gatys et al. (2015) developed a technique to synthesise textures using a model based on correlations in the feature space of a pre-trained convolutional neural network, namely VGG-19. Our research expands upon this foundation by exploring the method's adaptability to a different CNN architecture, demonstrating its efficacy with ResNet34. We also introduce a gradient threshold as a stopping criterion for the synthesis process, significantly enhancing computational efficiency without compromising texture quality. Further, we investigate various seed image types, especially in terms of their frequency domain content, to determine their impact on the synthesis outcome. Additionally, our study examines the relationship between scale invariance in textures and synthesis quality, utilising wavelet transform for a multi-scale analysis.

Item Type:	Thesis (Bachelor's Thesis)
Supervisor name:	Tursun, O.T.
Degree programme:	Computing Science
Thesis type:	Bachelor's Thesis
Language:	English
Date Deposited:	21 Dec 2023 09:32
Last Modified:	21 Dec 2023 09:33
URI:	https://fse.studenttheses.ub.rug.nl/id/eprint/31763

Actions (login required)

View Item