The Study Of Image Segmentation And Matching For Gastric Epithelial Cells

So far, the development of Medical image Matching and Segmentation has wide range ofrelated algorithms, but it still does not fully satisfy our practical needs. With the continuousdevelopment of new theories, new methods of image matching and segmentation areconstantly emerging, But in the practical image projects, that with an algorithm that canachieve the desired results can not be any kind of single image segmentation algorithm.It isdifficult for all of the images are relatively satisfied with the results obtained. Differentimages adopt different methods, Then we can get a better result of matching andsegmentation.This dessertation has a background of the project of Jiangxi Science and TechnologyDepartment < Development of Software for pathological analysis of tumors> .for the cellstructure of gastric epithelial, shape, sparse level, arranged in the shape, between differentnucleuses, between the nucleus and the duct, phenomenon of adhesion between the duct andthe duct is very serious. Studying the image matching and segmentation techniques of gastricepithelial cells, we mainly do the following work:Firstly, using SIFT algorithm effectively in gastric epithelial cell image matching, canassist doctors to remove or find similar patients in different regions of images, fasten imagesegmentation and recognition; segmentation can not be done if no image with clear orrelatively poor light conditions is found, improve the segmentation efficiency; also can findthe same clear image of patients is relatively clear image area division operation, the doctorsprovide more effective analysis of data; and find the results of images in different periods ofthe same patients and then analyze them.Secondly, pre-denoise the matched images. And using this improved Mean Shiftalgorithm of Segmentation: The Gaussian Kernel Mean Shift Algorithm for the kernelfunction, take the unit matrix as the bandwidth matrix through pretreated gastric epithelial cellimage segmentation, through qualitative segmentation evaluation methods to determine themethod.it improves the traditional Mean Shift Algorithm and watershed segmentationalgorithm such as over-segmentation problem, the unit matrix is the bandwidth matrix, itimproves the operating efficiency of the program. And can be faster and better distinguish thenucleus, cytoplasm and background regions.it is more suitable to Gastric epithelial cells forimage segmentation.Thirdly, using the algorithm of image tracking, we get image of target area byquantitative data segmentation. Then we set a two-dimensional linked list, the main track listtrace the land area and Vice list tracking cell area. The better images are divided into a continuous closed cytoplasm area, nucleus area, glands and background region, and thenucleus and cytoplasm are saved as two separate two-dimensional list, it saves the memory,improves data retrieval efficiency of the cell , once again provides convenience. for the searchof the cell and cell characteristic parametersFinally, This dissertation applies the algorithms of SIFT-based matching operator, basedon improved Mean Shift and tracking algorithm into the software of diagnosis of cancerpathology analysis, and develops a relatively complete pathological analysis and informationmanagement system.