Non-parametric Model-based Image Segmentation Method

So-called image segmentation is the technology and processing to extracts the interested target from the integral image. It is the foundation of understanding image and it is also the basic premise of image visual analysis and pattern recognition. At the same time, it is also a classic problem .So far, there is not an image segmentation algorithm suitable for all images. Therefore, the purpose of designing more generic, efficiency and accuracy algorithm become more popular in some scientists.Level set method is a new method for solving geometric curve evolution, it can naturally handles the changes of curve topology. It expressing the closed curve or closed surface in a hidden way, it can evolution the curves by numerical solution of partial differential equations.At the beginning of this thesis, we studied the classic seed-point algorithm, and present an improved seed-point algorithm. This new algorithm can process RGB image directly without changing color space. However, the segmentation results of seed-point algorithm are highly dependent on the selection of parameters and not conducive to automated segmentation. So, in this thesis we mostly focuses on a non-parametric algorithm for image segmentation.Non-parametric segmentation algorithm is an efficient technique for image segmentation. In this method, we introduced the information theory into the level-set method. We convert the segmentation problem to maximize the mutual information of regional labels and image pixel. This algorithm can automatically extract interested target from the integral image without human intervention and does not require any pretreatment before segmentation. This method can process more image types than the traditional parameter methods.