Texture classification using an invariant texture representation and a tree matching kernel
by Somkid Soottitantawat
Title: | Texture classification using an invariant texture representation and a tree matching kernel |
Author(s): | Somkid Soottitantawat |
Advisor: | Surapong Auwatanamongkol, advisor |
Degree name: | Doctor of Philosophy |
Degree level: | Doctoral |
Degree discipline: | Computer Science |
Degree department: | School of Applied Statistics |
Degree grantor: | National Institute of Development Administration |
Issued date: | 2010 |
Digital Object Identifier (DOI): | 10.14457/NIDA.the.2010.54 |
Publisher: | National Institute of Development Administration |
Abstract: |
The real world is rich in many textures, which can be regarded as the visual appearance of surfaces. They may be perceived as being smooth or rough, coarse or fine, homogeneous or non-homogeneous, etc. Moreover, textures within real images vary in scale, rotation and illumination. Several researchers have proposed texture analysis methods to describe textures in many applications, such as computer vision, pattern recognition, image retrieval, scene image analysis, and so on. Although the analysis of texture properties has attracted the interest of researchers for more than three decades, in this dissertation, an alternative approach for texture classification using an invariant texture representation, tree-of-keypoints and a tree matching kernel is proposed. The approach identifies regions of a given texture image using Speed-Up Robust Feature or SURF descriptors. The regions of the training texture images are then clustered into a tree of non-uniformly shaped regions based on their distribution using a hierarchical k-means algorithm. The tree structure forms a tree of key points which are used to determine the similarities between two texture images. A similarity is computed based on an approximate matching kernel called a tree matching kernel. Finally, Support Vector Machines (SVMs) that utilize the tree matching kernel are constructed to classify textures. The performance of the proposed method was evaluated through experiments performed on textures from the Brodatz and UIUCTex datasets and, in all experiments with the three learning schemes and different three weighting schemes, performed consistently better than other previously reported methods. |
Description: |
Thesis (Ph.D. (Computer Science))--National Institute of Development Administration, 2010 |
Subject(s): | Visual texture recognition
Pattern recognition systems Optical pattern recognition Support vector machines Kernel functions |
Resource type: | Dissertation |
Extent: | xxi, 233 leaves : ill. ; 30 cm. |
Type: | Text |
File type: | application/pdf |
Language: | eng |
Rights: | This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License. |
URI: | http://repository.nida.ac.th/handle/662723737/282 |
Files in this item (CONTENT) |
|
View ทรัพยากรสารสนเทศทั้งหมดในคลังปัญญา ใช้เพื่อประโยชน์ทางการเรียนการสอนและการค้นคว้าเท่านั้น และต้องมีการอ้างอิงแหล่งที่มาทุกครั้งที่นำไปใช้ ห้ามดัดแปลงเนื้อหา และทำสำเนาต่อ รวมถึงไม่ให้อนุญาตนำไปใช้ประโยชน์เพื่อการค้า ไม่ว่ากรณีใด ๆ ทั้งสิ้น
|
This item appears in the following Collection(s) |
|
|