Ransac Based Motion Compensated Restoration for Colonoscopy Images

Abstract

Colonoscopy is a procedure that has been used widely to detect the abnormality in a colon. Colonoscopy images suffer from a lot of problems that make it hard for the doctor to investigate/ understand a colon patient. Unfortunately, with the current technology, three is no way for doctors to know if the whole colon surface has been investigated or not. We have developed a method that utilizes RANSAC-based image registration to align sequences of any length in the colonoscopy video and restores each frame of the video using information from these aligned images. We proposed two methods. First method used the deep neural net for the classification of informative and non-informative image. The classification result was used as a preprocessing for alignment method. Also, we proposed a visualization structure for the classification results. The second method used the alignment to decide/classify the bad and good alignment by using two factors. The first factor is the accumulated error and the second factor contain three checking steps that check the pair error alignment beside the geometry transform status. The second method was able to align long sequences.