Title

Feature Relational Mapping for Remote Sensing Images Using Directional Edge Analytical Models

Author

Junfeng Gu

Date of Award

2007

Degree Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Computing

First Advisor

Dia Ali

Advisor Department

Computing

Abstract

An integrated approach of feature analysis--Directional Edge Analytical Model (DEAM) is presented. Remote sensing and image processing with feature extraction are significant for collecting information. Many different techniques, invariant to different types of geometric transformations, contextual conditions and task demands, have been developed for recognizing features of objects. Most of current feature analysis has emphasized non-remote objects and has lacked handling with the properties of remote sensing images. In addition, Relational structures are particularly suitable for representing complex features and for providing matching solutions; this area of research is still in very early stage. Using directional edges as descriptors for relational features, DEAM provides supervised models with matching procedures. Novel methods with adaptive threshold detectors and directional descriptors are designed according to the properties of remote sensing images. DEAM analyzes the connection of the functional edges in an object, and then it builds the analytical model by feature descriptors. It utilizes "many to many relations" with "relation of relations" to reduce the redundancy and to increase the accuracy of a training set. After feature relational mapping, the corresponding descriptors between the model and the target will assist researchers in identifying the object. DEAM demonstrate superiority in quantifying structural relationships and in normalized features that are desired for many analytical applications.