Date of Graduation

8-2019

Document Type

Dissertation

Degree Name

Doctor of Philosophy in Computer Science (PhD)

Degree Level

Graduate

Department

Computer Science & Computer Engineering

Advisor/Mentor

Gauch, John M.

Committee Member

Gauch, Susan E.

Second Committee Member

Andrews, David L.

Third Committee Member

Cothren, Jackson D.

Keywords

colon; deep neural net; medical image processing; motiion compersated; Non linear motion compensated; RANSAC

Abstract

Colonoscopy is a form of endoscopy because it uses colonoscopy device to help the doctor to understand a colon patient. Enhancing the quality of Colonoscopy images is a challenge because of the wet and dynamic environment inside the colon causes many problems even the colonoscope devise has a good quality. Some of these problems are blurriness, specular highlights shiny areas.

In this work, different kinds of techniques have been investigated in order to improve the quality of colonoscopy images. Also, variety of preprocessing approaches (removing bad images, resizing images, median filtration with and without image resizing) have been conducted to aid the automatic process of alignment/registration methods for colonoscopy images.

For example, removing bad images using different kinds of classifier helps our approach that involve using RANSAC method to work automatically and remove some the specular highlight and some shiny spots in the colonoscopy images. It also helps a lot to reduce the time required for the implementation. On the other hand, considering median and resizing images did not help a lot. Also, a variety of visualizing techniques have been suggested and used to help doctor to visualize the colon images.

Another technique we used to improve the quality of colon images is the motion compensation with temporal filter. In this experiment, we did not use any kind of the mentioned preprocessing methods. Instead, we used the good and bad images and our approach helped to remove some of the problems that the colon images suffer from such as specular highlights and blurriness. We concluded from our experiments that our second approach which is the motion compensation with temporal median filter is more effective in removing specular highlights than RANSAC method.

Download

Share

COinS