COMING SOON! PQDT Open is getting a new home!

ProQuest Open Access Dissertations & Theses will remain freely available as part of a new and enhanced search experience at

Questions? Please refer to this FAQ.

Dissertation/Thesis Abstract

Remote Sensing Image Segmentation and Object Extraction Based on Spectral and Texture Information
by Yuan, Jiangye, Ph.D., The Ohio State University, 2012, 110; 10631406
Abstract (Summary)

The increasing availability of remote sensing data and the growing demand on geoinformation for various kinds of spatial management issues have catalyzed the development of new methods to analyze and understand collected data more effectively and efficiently. This dissertation investigates two critical tasks in remote sensing data analysis, image segmentation and object extraction based on exploiting spectral and texture information.

Locally excitatory globally inhibitory oscillator network (LEGION) provides a general framework that is capable of segmenting images and extracting objects of interest. We incorporate spectral information from multiple bands into LEGION networks, which are applied to extracting seagrass patches from hyperspectral data. Based on the LEGION framework, we develop a new automatic road extraction method using the medial axis transform and alignment-dependent connections. The evaluation using a road extraction benchmark dataset shows that our method produces improved results.

We propose a new texture segmentation method. Local spectral histograms are feature vectors consisting of histograms of chosen filter responses, which capture both texture and nontexture information. For an N-pixel image, we construct an M×N feature matrix using M-dimensional feature vectors. Based on the observation that each feature can be approximated through a linear combination of several representative features, we express the feature matrix as a product of two matrices – one consisting of the representative features, and the other containing the weights of representative features at each pixel used for linear combination. When representative features are manually given, the segmentation result is obtained by least squares estimation. With unknown representative features, we utilize singular value decomposition and nonnegative matrix factorization to factor the feature matrix, which leads to segmentation results. The scale issue is also investigated, and an algorithm is presented to automatically select proper scales. This algorithm does not require segmentation at multiple scale levels.

We apply the proposed segmentation method to combined spectral-texture segmentation for remote sensing images. Local spectral histograms can capture spectral and texture information by applying different filters to spectral bands. The proposed method integrates the information to produce segmentation.

We conduct experiments on texture and natural image datasets to show the effectiveness of our approach. To evaluate the performance of the method on combined spectral and texture segmentation, we test our method on IKONOS panchromatic/multispectral bundled images. The comparison with other methods demonstrates that the proposed method improves segmentation accuracy by 14% on the testing images.

Indexing (document details)
Advisor: Li, Rongxing, Wang, DeLiang
Commitee: Yilmaz, Alper
School: The Ohio State University
Department: Geodetic Science and Surveying
School Location: United States -- Ohio
Source: DAI-B 78/11(E), Dissertation Abstracts International
Subjects: Geographic information science, Computer science
Keywords: Extraction, Information, Object, Remote, Segmentation, Texture
Publication Number: 10631406
ISBN: 978-0-355-01649-9
Copyright © 2021 ProQuest LLC. All rights reserved. Terms and Conditions Privacy Policy Cookie Policy