Image Segmentation Research Of Abdominal Macrophages In Mice

As one of the most important applications of image segmentation, segmentationof cell images obtains extensive concerns of many researchers in the area of computerimage processing. The segmentation methods of cell images have been developedrapidly with the development of communication technology, information technologyand computer technology and become the one-up research task in present imageprocessing domain. In this paper, we firstly retrospect the definition and classificationof segmentation, then introduce some traditional and new segmentation methods indetails and finally, the application in cell image segmentation.The watershed algorithm is a kind of mathematical morphologic imagesegmentation methods. It could get the precise edge which is continuous, closed andof the width of single-pixel. The main limitation of watershed transform is theover-segmentation due to its sensitivity to noises, even a little of noise will lead to alot of scattered and meaningless regions. To overcome this difficulty, in this paper, weadopt watershed arithmetic based on distance transformation. Firstly, the image isthresholded by means of the well known Otsu's thresholding algorithm, then thedistance transform is carried out to change the position information into the grayinformation, whereafter, edge points are marked by using watershed algorithm andthen the cell clustering is completed.The level set method based on geometric deformable model, which translates theproblem of evolution of 2-D (3-D) close curve (surface) into the evolution of level setfunction in the space with higher dimension. However, as the traditional level setmethod uses just the local marginal information of the image, it is difficult to obtain aperfect result when the region has a fuzzy or discrete boundary, and the leakingproblem was inescapable appeared. In this paper, we introduce Mumford-Shad modelfor image segmentation proposed by Chan and Vese. The stopping function of thismethod does not depend on the local gradient. Hence, an ideal segmentation resultcould be obtained even if the boundary is fuzzy or not continuous.