Perception-Motivated High-Quality Stylization

It appears your Web browser is not configured to display PDF files. Download adobe Acrobat or click here to download the PDF file.

Click here to download the PDF file.

Creator: 

Li, Hua

Date: 

2014

Abstract: 

Non-photorealistic rendering (NPR) algorithms are used to produce stylized images, e.g., in a painted or stippled style. This thesis presents a new family of automatic methods for effects and styles including halftoning, screening, stippling, line art, and mosaics. Our proposed algorithms are motivated by perceptual effects including contrast and similarity to present structure preservation. Meanwhile, the novelty also shows in exploring primitives in distribution for our NPR algorithms.

First, we propose a novel contrast-aware error diffusion algorithm based on a priority-based scheme,
called Prioritized Contrast-based Error Diffusion (CED), to generate stylized imagery. The core idea of this technique is to implicitly and progressively preserve the original tendency in contrast. This method generates halftoning with a success in similarities for tone, structure, and contrast, and with good visual appearance. We extend Prioritized CED with variations on masks and priority configuration to create screening style with non-uniform patterns. To apply Prioritized CED to stippling generation, we design a new modification to the contrast-aware error distribution to provide
density control in dot distribution. A set of varied stippling styles are produced. Thanks to the good structure preservation, all resulting effects are quite good. We also propose a technique of pixel clustering and employ skeletonization for generation of line art with long strokes. Pixel clustering produces long lines by processing a set of pixels, which align along the edge guidance from a smoothed bilateral filter on an edge tangent field (ETF). Skeletonization captures large-scale structure of an image. We can create simplified line art with clean and elegant visual appearance.

Second, we propose a novel automatic approach to construct an artistic tessellation (AT) through the growth of curves in a particle system. The core idea is to present similarity, and in addition, to consider spatial balance during curve growth. The AT method is applied to create natural and abstract patterns. The further exploration guided by a bilateral smoothed ETF from an image can simulate stained-glass mosaics.

Subject: 

Computer science

Language: 

English

Publisher: 

Carleton University

Thesis Degree Name: 

Doctor of Philosophy: 
Ph.D.

Thesis Degree Level: 

Doctoral

Thesis Degree Discipline: 

Computer Science

Parent Collection: 

Theses and Dissertations

Items in CURVE are protected by copyright, with all rights reserved, unless otherwise indicated. They are made available with permission from the author(s).