Level Set Image Segmentation Method Combined With The Shape Of A Priori

Image segmentation is to separate an image into certain un-overlapped homogeneity regions which utilize certain characters of image, such as gray level, color, texture and edge information of the image. Image segmentation have a critical influence on later image analysis and image understanding. Therefore, the research of image segmentation methods has both the academic and application value.Active contour model can combine the image of low-level visual such as gray, edge, texture properties with the image target prior shape, color knowledge so on using a method that consistent with human vision. As the level set method appeared, which greatly promoted the development of active contour model. Geometric active contour model theory of curve evolution and level set method more greatly expanded the scope of application of the active contour model. Because the image always have noising, missing and boundary ambiguity, geometric active contour model can not obtain good segmentation results. It’s important to fuse shape prior information into the level set segmentation method.This paper mainly takes the image of shape prior information and the level set model and do the three aspects of the work:(1)This papper expounds the basic theory of curve evolution and level set method.Introduce the geometric active contour model and several classical models (CV model, LBF model and LGIF model). Introduced the vector CV model, and the vector CV model is improved in two aspects. First of all, combined with Li’s idea that adding distance penalty term into the vector CV model to solve the issue of recurve in the process of evolution; secondly, given a new termination condition of curve evolution.(2) Chan et al proposed the vector CV model to solve the problem that the traditional CV model cannot segment the vector images. But it has a bad effect on the complex images that have noise or occlusions, so the paper proposed the vector CV model combining shape prior. Its energy functional is mainly composed of shape prior information term and image area information term and distance regularization term. When the evolved active contour and shape prior have similar positions, the contour stops evolution. According to the affine transformation of shape, using a gradient descent algorithm for template to match, which makes the algorithm more flexible.The model has good segmentation result for the noise and clutter image, and eliminates the need of reinitializing when the curve evolves.(3) the method by the definition of the level set function and shape prior distance will shape prior information is added to the vector CV model to guide the segmentation, however, this method is only suitable for the single prior condition, the practical application is generally not possible to accurately get the shape prior target segmentation. This paper combined the Tsai et al proposed based on kernel principal component analysis (Kernel Principle Component Analysis, KPCA) segmentation method with multiple shape priors vector CV model image. KPCA can express better shape priori knowledge and allows nonlinear transformation or a quite difference between the object and the prior shape. Moreover,our segmentation model includes the image term and the shape term to balance the influence of the global imageinformation and the shape priori knowledge in proceed of segmentation. The comparative results show that KPCA can more accurately identify the object with large deformation.