Research And Implemention Of Image Parsing Method Based On Machine Learning

For medicine, in the process of the diagnosis of gastrointestinal disease will produce alarge number of image data, traditional diagnostic methods mainly rely on the clinicians’manual analysis, it is time-consuming. Therefore, the computer-aided diagnosis methodbased on image analysis technology has been more and more attention. In disease ofdigestive tract, stomach disease has a high incidence rate. At present, commonly useddiagnostic methods has capsule endoscopy and gastroscope. For the defect of its strcture,capsule endoscopy can not provide enough diagnosis basis in the process of stomachdisease diagnosis, thus gastroscope is more widely used. Nevertheless, the research aboutcomputer aided diagnosis according to the gastroscope image is relative lacking.This thesis research and implementation of a kind of image analysis method forgastroscope image, that based on machine learning. Based on interactive proposed animage segmentation method, and use it to label the gastroscope images. Complete sampledata collection and the establishment of the sample database on labeled images. Then,extract color and texture feature descriptor from the samples, analysis and research eachcharacteristics descriptor so as to choose the good performance features. Combine withthese features, using SVM classifier to realize the sample data training, and create thetemplate, under the action of the template, achieve the purpose of detecting lesions. Thesuperpixel segmentation method was used to divide sample area, and has a good effect.Finally, on the basis of realizing each technology link, we completed the lesions diagnosissystem for gastroscope image. And the accuracy of the image analysis method wasvalidated through experiment, which explains the feasibility and effectiveness of themethod.