Reseach On Image Segmentation Based On Active Contour Model

Image segmentation vision has been an important research topic in the field of computervision, it aims to extract from the background for target objects which was needed. In thesemantic level, the segmentation of the target can be a medical image in organ, tissue;invaders from a surveillance video; mechanical moving parts in robotics or nautical andmilitary goal. Visible image segmentation as a basic discipline field, has played a veryimportant role in many practical applications. In actual images, there are noise, weak edgesand other issues which bring great obstacles to the traditional image segmentation method.And because the active contour model can get sub-pixel accuracy and close, smooth contourcurve, has been widely used in the field of image segmentation.Active contour model based on different features can be divided into two categories:active contour model based on edge and active contour model based on region. The formeruses the gradient information of the image to construct edge stopping function (ESF), so thatthe curve evolution stops at the edges of the image, such as a geodesic active contour model(GAC); the latter is the use of image statistical information to control the evolution of curve,which has a better treatment effect for the noise and weak edge. Our model has the advantagesof both of them, which based on the classic GAC model, and it was embedded with the localgray fitting information, so our model formed a new method which can handle non-uniformgray image.Firstly, because of the introduction of Signed Pressure Function(SPF), our method hasthe very good processing effect for weak and blurred edges; secondly, because which we useis local gray fitting statistical information, so this method has a certain processing capacity forthe non-uniform gray, and it also has a certain robustness for the noise, and because weproposed to estimate the local fitting value by Gauss mixture model (GMM) and EMalgorithm, so the estimated result is more accurate;meanwhile, our model is based on thegeneralization of the GAC model, it inherited the advantages of the active contour modelwhich was based on the edge, and it can use the appropriate initial conditions to specify thetarget which was needed to be segmented. Compared with other active contour models whichwere embedded local information such as LBF model, our method is more robust to initialconditions, and in different initial conditions our method can converge to the same approximate results. Through experiments, we also demonstrate that the proposed method hadcharacteristics and advantages which compared with other method.