Abstract

A feature-selection technique based on measures of global class separability in multidimensional feature space is proposed for classifying monochrome digitized imagery by machine. Feature-selection procedures are an essential step in optimal classification in reduced feature space. Textural features constitute the type of measurements used to characterize image data due to its monochrome nature. IV The ability of the proposed feature-selection technique to provide an optimal environment for classifying image pixels is measured by the Gaussian Maximum Likelihood method. The appropriateness of using textural features to characterize monochrome digital image data is assessed in similar fashion. The robustness of the proposed feature selection technique, and that of use of textural features, to provide for accurate and effective image processing is tested by analyzing several monochromatic images which contain multiple ground-cover classes, various resolutions, orientations, grey-level quantization levels, and individual textural feature parameter settings.

Library of Congress Subject Headings

Image processing--Digital techniques; Remote sensing--Data processing

Publication Date

5-20-1989

Document Type

Thesis

Department, Program, or Center

Chester F. Carlson Center for Imaging Science (COS)

Advisor

Schott, John

Comments

Note: imported from RIT’s Digital Media Library running on DSpace to RIT Scholar Works. Physical copy available through RIT's The Wallace Library at: G70.4.R624 1989

Campus

RIT – Main Campus

Download

Share

COinS