Term of Award

Spring 2012

Degree Name

Master of Science in Computer Science (M.S.)

Document Type and Release Option

Thesis (open access)

Copyright Statement / License for Reuse

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.

Department

Department of Computer Sciences

Committee Chair

Robert Cook

Committee Member 1

Lixin Li

Committee Member 2

Vladan Jovanovic

Committee Member 3

James Harris

Committee Member 4

Nathan Yanasak

Abstract

Computer vision is a field of computer science that includes methods for acquiring, processing, and analyzing images. Image registration is one of the methods used in the computer vision field to transform different sets of data into one coordinate system to align images. Registration is important in order to be able to compare or integrate the data obtained from multiple measurements. Rigid image alignment is a type of image registration technique used to align two two-dimensional images into a common coordinate system based on two transformation parameters, translation and rotation. Before any comparative studies can be performed on two images acquired at different times, it is crucial to align the two images for correct processing later on. In our research study, we are analyzing the accuracy of registering images using two rigid image alignment algorithms, namely the Principal Axes algorithm and the Fast Fourier Transform (FFT) based phase correlation algorithm. The software for registering images using these two methods is written in MATLAB R2011a. We also compared our results with alignments achieved for the same images using an existing Statistical Parametric Mapping (SPM8) package for registration. Image registration algorithms have been used in many applications and accordingly, algorithms are adopted to suit a particular application. Images used for registration can be derived from different capturing devices like camera, scanner, satellite sensors, etc. Our registration software is based on work with images acquired from a Magnetic Resonance Imaging (MRI) scanner and especially for images taken of a quality assurance (QA) phantom. A QA phantom is used to test the quality of images acquired by measuring different QA parameters on images acquired over a period of time. Images acquired from the MRI scanner at different times are geometrically transformed by rotation and translation. In practice, the maximum angle by which the phantom will get rotated at different times due to varying positioning in the scanner will not be greater than 50 degrees and the maximum displacement will always be less than 50 pixels based on our experience while scanning. By comparing future phantom images with the first image in the series, we can perform a series of Quality Assurance steps to measure any degradation in the MRI device. The QA results can then be used to apply inverse transformations to new customer images to improve their quality. The first step in the QA process is image registration, which is the topic of this thesis. To test the implementations, we rotated and translated known images then we applied the two algorithms and compared the results to the known translation and rotation values. Our analysis shows that the Principal Axes method could successfully register 17 of the 22 non-aligned test images, the FFT method registered 21 test images successfully whereas SPM8 with default settings showed correct alignments for only 9 images in our case study as per our requirement. The Principal Axes algorithm performed better image alignment when the two images were displaced by a larger distance, and the FFT based algorithm performed better for larger rotation angle differences among images. Hence, we conclude that our algorithms have the potential for inclusion in the new QA process.

Research Data and Supplementary Material

No

Download

Share

COinS