Date of Award

1998

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Computer Science

First Advisor

John M. Tyler

Abstract

Registration is the process of spatially aligning two objects and is normally a preprocessing step in most object recognition algorithms. Registration of images and recognition of signatures of objects in images is important for clinical and diagnostic purposes in medicine. Recognizing structure, potential targets for defense purposes and changes in the terrain, from aerial surveillance images and SAR images is the focus of extensive research and development today. Automatic Target Recognition is becoming increasingly important as the defense systems and armament technology move to use smarter munitions. Registration of images is a preprocessing step in any kind of machine vision for robots, object recognition in general, etc. Registration is also important for tuning instruments dealing with images. Most of the available methods of registration today are operator assisted. The state of registration today is more art than science and there are no standards for measuring or validating registration procedures. This dissertation provides a viable method to automatically register images of rigid bodies. It provides a method to register CT and MRI images of the brain. It uses wavelets to determine sharp edges. Wavelets are oscillatory functions with compact support. The Wavelet Modulus Maxima singularides. It also provides a mechanism to characterize the singularities in the images using Lipschitz exponents. This research provides a procedure to register images which is computationally efficient. The algorithms and techniques are general enough to be applicable to other application domains. The discussion in this dissertation includes an introduction to wavelets and time frequency analysis, results on MRI data, a discussion on the limitations, and certain requirements for the procedure to work. This dissertation also tracks the movement of edges across scales when a wavelet algorithm is used and provides a formula for this edge movement. As part of this research a registration classification schematic was developed.

ISBN

9780591998061

Pages

204

Share

COinS