| Image segmentation is the basic step to pattern recognition, target tracking, target reconstruction, image registration, and one of key techniques during the field of image information. At the same time, it occupies the inevitable position in the field of image processing. Because of the complexity of image scene, image segmentation always is a difficulty and a focus of research in all related areas of images.In the various methods of image segmentation, active contour model based on level set because of its unique characteristics has attracted the favor of the researchers. Segmentation method based on level set is still in development. But there are many problems such as slow rate of segmentation, huge computation cost, sensativity to noise etc.The image segmentation method based on level set active contour model is studied and discussed in this paper. Firstly, we summarize the situation of image segmentation, and make a detailed description of the mathematical basis and theories about level set. Secondly, the several classic active contour model based on level set is analyzed, and then the LBF model is improved. Active contour model driven by similar degree and local fitting energy is presented. Finally the entropy of the image is introduced and a level set model of two stage segmentation is presented. This thesis mainly has made the following research:(1) To overcome the disability of LBF model for segmentation on texture image, such as poor robustness for noise image, slow speed of segmentation, this paper proposes a novel active contour model driven by similar degree and local fitting energy. The energy term which is from the difference of intensity distribution between the inner and outer contour region is introduced into LBF model. At the same time, the gradient information of the image is used to define an acceleration factor to weighted the model. Because the sigmoid function in the form is adopted, the noise amplification is suppressed. At the same time, the gray distribution probability model itself is insensitive to the noise. Therefore, the improved model has strong robustness to noise and can efficiently segmenting texture images, with fast segmentation speed. The experiment result demonstrates the analysis.(2) To overcome the disadvantage that the mixed model ineffectively trades off between the global forces and local forces. Based on the image entropy, we present a two-stage level set segmentation model. Because the image entropy changes during the contour evolution, it taken as a two phase of the indicator function. At the same time, we combine the feature of the CV model that has fast segmentation speed and is insensitivitive to initial contour and the feature of LGDF model that has the precise segmentation. The presented method achieves the better segmentation result. And the effectiveness of the method is verified by experimental results. |
