Reasearch And System Realization Of Wireless Capsule Endoscopic Image Retrieval And Video Abstract Method

Wireless capsule endoscope is a milestone in the digestive tract detection technology history,which not only overcome the defect about the traditional gastrointestinal detection methods for human injury,but also make it come true to detect the entire small intestine.Since wireless capsule endoscopy has come,there are now more than one million clinical cases in record,and gradually become the first choice in the field of gastrointestinal detection methods.Wireless capsule endoscope achieves the entire digestive tract shooting by the use of human intestinal peristalsis after swallowed.Typically,the entire inspection process takes 8 hours and takes about 30000 to 60000 digestive tract images.However,screening out lesion images from these digestive images for disease diagnosis is a boring job for clinical physicians.Therefore,wireless capsule endoscopic computer-assisted diagnosis and treatment technology has forested in order to effectively reduce the labor intensity of doctors and improve diagnostic efficiency.In this thesis,we focus on two aspects(image retrieval and video abstract)about wireless capsule endoscopy images.Image retrieval is designed to quickly select a similar image from a large database about wireless capsule endoscopic image,to reduce the time clinician looks for a similar image.Video abstract is reduce the total number of frames of wireless capsule endoscopic image sequence,and improve the diagnostic efficiency.About image retrieval technology,this thesis proposes a content-based image retrieval method for wireless capsule endoscopic images based on the characteristics of wireless capsule endoscopic image.About video abstract technology,two methods are presented in this thesis,which are saliency based wireless capsule endoscopy video abstract and convolutional neutral network based wireless capsule endoscopy video abstract.In this thesis,the color feature,texture feature and Sift feature are extracted to describe the characteristics of wireless capsule endoscopic images,which the color and texture features are fused to describe the sensory characteristics of human eye,and Sift feature is used to describe the local detail information of the image.Finally,the retrieval results are obtained according to the similarity by the ratio method.The experimental results have shown that the content-based wireless capsule endoscopic image retrieval technology can effectively retrieve similar images with an average retrieval accuracy of 88.3%.About saliency based wireless capsule endoscopy video abstract method,this thesis improves the traditional Itti's salient map extraction method,and makes it more suitable for representing the wireless capsule endoscopic images.Based on the extracted salient map,the color feature and texture feature are extracted and fused to describe the wireless capsule endoscopic images.Then,the keyframes detection is utilized by the mutational detection and gradient detection methods,which avoids missing the gradient images in the video.The video abstract can be obtained by the keyframes combination.The sensitivity,specificity,accuracy and compression ratio of the experiments have achieved good results,which proves that this method can provide a reference method for wireless capsule endoscopic video abstract.In this thesis,the videos are divided into time series according to the characteristics of wireless capsule endoscopic videos.Then,the K-means clustering method to the time series images is carried out to remove the similarity images in the series,and the residual images are calibrated by the professional doctors.After that,the AlexNet convolutional neural network is used to train the keyframes recognition model.Finally,the combined keyframes are formed to generate the wireless capsule endoscopic video abstract.The experimental results have shown that this method not only has good experimental performance,but also can effectively preserve the lesion information in the video.According to the above research contents,this thesis uses the MFC framework to realize the wireless capsule endoscopic image retrieval system and video abstract system,which is aimed at the convenient experiment and operation to the users.Finally,this thesis summarizes and analyzes the overall work,and prospects the follow-up work.