Image Segmentation Based On Curve Evolution Theory

Image segmentation technology is widely used in many fields of computer vision andcomputer science, it is a key technology in the image engineering system, and which is aprerequisite for high-level image analysis and image understanding. The so-called imagesegmentation is to divide image into regions and the segmentation of the target object.In recent years, the active contour model based on curve evolution theory is a type of imagesegmentation algorithm, which is widely concerned by domestic and foreign researchers becauseof the attribute that their geometry expressions, physics equations and approximation theory canbe unified in the same contour extraction process. The content of this paper is focused on theSnake model which is the parameter active contour model and Chan-Vese model (C-V model)which is the geometric active contour model, the research is expanded around the models in theapplication of image segmentation, aimed at the shortage of the models, the major work andinnovations in paper are as follows:(1) Aimed at sensitivity to initial contour and lack of curvature constraint in the formulationof the directional Snake functional, an automatic contour segmentation algorithm based onimproved watershed transformation and active contour model is presented. First, a modifiedwatershed algorithm based on the marker function and the mandatory minimum technology isproposed in this paper to deal with the over-segmentation. Then the improved watershedalgorithm is adopted for pre-segmentation, the extracted object contour is taken as initial contourof the Snake model. Finally, an external force which is related to the curve shape is added in theSnake model for making up the lack of curvature constraint in the formulation of the energyfunctional. The improved Snake model can achieve good results in the liver image recognitionand segmentation when applied to the MR images of the abdomen.(2) C-V model is a classic approach to segment image based on variational level set. It isespecially helpful for medical images analysis, which has complex topology constructions, strongnoises and lower contrast. But C-V model depends on initial contour. Image thresholding is also asimple and popular image segmentation method, which separates an image into two parts, andgets the target sub-image from background image. In the article, by analyzing the distribution ofimage gray level histogram, the authors propose a new method which initializes contour usingthreshold. The novel C-V model used in dental plaque image segmentation and experimentsillustrates that the method proposed can help segment images. It provides a preliminary work tofurther analyze dental plaque. (3) Taking into account the problem that level set optimization method in C-V model isdifficult to determine the best iteration times and easy to fall into the local optimum, butconsidering that graph cuts algorithm can get the global optimum in a short time, a new iterationterminated algorithm based on graph cuts and single level set is proposed. First of all, target areasets an initial contour, and iterates the contour by the C-V model without re-initialization, whenthe change rate of area within the contour is less than the given threshold, terminate the iteration.Then this contour is taken as the initial contour of the graph cuts algorithm for imagesegmentation. The experiments results show that this method greatly reduces the iteration time,has a higher robustness and a better effect for image segmentation compared to the original C-Vmodel.