Research On Reduction Of Wireless Capsule Endoscopy Image Sequence

Wireless Capsule Endoscopy (WCE) is a novel technology in first10years of this century. WCE is mainly used for investigating the diseases and abnormalities in small intestine with a noninvasive manner which overcomes the problem of traditional Capsule Endoscopy which can not examine the diseases of small intestine. Although WCE is used in clinical examination only10years, the many famous hospitals in the world have chosen the WCE as a first device for examine the diseases and abnormalities in small intestine. WCE takes about6-8hour to traverse through the entire digestive tract and captures approximately50,000images of gastrointestinal wall scene. Currently, the major weakness of WCE examination is that it takes a long time to examine the whole WCE video, which is time-consuming, intense labor and tedious for WCE clinician. Therefore, the Computer Aided Diagnosis (CAD) for WCE images processing has became a research focus gradually. The goal is reduce the examination time and improve the results of disease diagnosis in small intestine.This dissertation focuses on estimating the similarity between images in WCE Image sequence and preserves the KeyFrame and representative images from WCE Image sequence based the similarity estimation, which reduce the number of WCE images in a sequence and the examination time. For achieving this goal, this dissertation proposes two WCE image sequence reduction algorithm:WCE image sequence reduction based on camera motion estimation (BAME-SIFTFlow) and WCE image sequence reduction based on loaction texture texton histogram (LTH).In the first algorithm, we propose WCE camera motion model and the method of motion estimation between two successive WCE images. Under this motion model, the imaging of WCE is subject to constraints, the first constraint come from the movement of WCE itself, and the second constraint come from the peristalsis of digestive tract. Based on this motion model, the WCE imaging motion is estimated in two stages. In the first stage of the coarse level estimation, the motion of WCE camera is estimated using the Bee Optimation Algorithm. In the second stage of the fine level estimation, the local gastrointestinal tract deformation base on the result of the first stage estimation is estimated using SIFT-Flow Algorithm. Then, the invalid region between successive WCE images is estimated using the backward and foreward motion estimation. The final we determine whether current image should be preserved as a key frame or not according the threshold of max-inscribed circle of the invalid region.In the second algorithm, we propose a similiarity estimation method between two successive WCE images with statistics learning. The similiarity estimation is achieved in two stages:learing and testing stages. In learing stage, the texture view set is extracted from sample image and the textons is built according to the texture view set. Then the location texton histogram of sample image is built with quadratic programming assignment scheme. In the testing stage, the location texton histogram of test image is built according the textons of sample image and the likelihood betweed sample image and test image is calculated. The final, we determine whether current image should be preserved as a key frame or not according the likelihood.From experimental results, we proposed two algorithms have higher recall (both74%) than ARPS and FCM-NMF algorithms. Especially, the LTH algorithm can estimate the similarity between two WCE images more effective than the BAME-SIFTFlow algorithm, the recall is77%. In additional, in sampling exprements, we found that our scheme is more consistent with the changes of WCE image sequence intensity than ARPS and FCM-NMF.Finally, the entire work is summarized and future work is prospected.