Multiphase Image Segmentation Based On Variational Level Set Methods

Multiphase image segmentation is one of the hot issues in the fields of image processing, computer vision etc. It has important applications in medical diagnoses, computer-aided surgery, machine vision, resource classification based on remote sensing and so on.Due to complexity of the problem, level set method with topologically self-adaptive capability becomes the mainstream method of multiphase image segmentation. On the other hand, variational method can conveniently combine different types of image information, such as edge, region, shape, texture and motion field, making it a better choice for building integrated image segmentation models with good universality and expansibility. In this paper, the mathematical models, numerical methods and applications of multiphase image segmentation based on variational level set methods are studied. The main content of the research work is:1) Previous research works in recent years on variational level set methods for image segmentation and related numerical difference methods are studied extensively and described in detail. Major problems for each method are analyzed.2) An improved model of traditional variational level set method for two phase image segmentation based on edge and region is proposed. In this new method, re-initialization of level set function is no longer needed. This method is also extended to accommodate multiphase image segmentation.3) A general region-based image segmentation model suitable for image with different noise distribution is established. The application domain for this method is later extended from piecewise constant image to piecewise smooth image.4) A general image segmentation model based on multiple level set functions that can be applied to images with an arbitrary number of phases is established. A new implementation strategy that can integrate aspects of different segmentation models through modularization is proposed. This new strategy has good extensibility and can be used to implement complex multiphase image segmentation operations.5) A new model for segmentation of multiphase image with noise and texture distribution is designed by combining multiphase image segmentation methods established above with image diffusion method. Image diffusion can reduce noise and transform textured regions into piecewise constant and piecewise smooth regions.6) For segmentation of piecewise smooth multiphase image, techniques for image inpainting are used to extend difference equation solution space into regular grid space. By doing so, the irregular boundary conditions of level set function evolution equation are eliminated.Finally, the proposed methods are tested and verified on some artificial images and a number of remote sensing image processing tasks, including coastal zone classification, coastline and watercourse extraction, acquisition of island vegetation distribution information and extraction of oil spill region.