A hybrid approach to problems in image processing

Digital image processing generally refers to the use of computer algorithms that deal with operations on or analysis of digital images. It covers a wide range of intriguing sub-topics. While these topics may vary from pattern recognition, classification to synthesis and generation, most techniques recognize the input as a raster image by default, treat it as a two-dimensional signal and applying standard signal processing techniques to it. In this paper, we attempt to solve specific problems of image segmentation, classification and enhancement by a hybrid approach. We perceive an input image not only as a 2D grid of pixels, but also as a connected graph, a collection of heterogeneous data points, or a surface mesh embedded in 3D space depending on the specific target problem. The broader framework makes it possible to combine image processing techniques with techniques that are typically applied to other research areas including digital geometry processing, machine learning, surface modeling and optimization. We demonstrate by experiments and comparisons that these methods, in concert with standard image operations such as filtering or feature detection, work particularly well and yield satisfactory results for the original target problems.

Name thesis:degree_name

Ph.D.

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Computer Science

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2010

Author dc:creator

• Xia, Tian

Contributors dc:contributor

• Yu, Yizhou
• Forsyth, David A.
• Hoiem, Derek W.
• Wang, Rui

• computer graphics
• image processing
• image vectorization
• image segmentation
• image classification

Statement dc:rights

• Copyright 2010 Tian Xia

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/16794

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/16794