Variational Level Set Methods For Multiphase Image Segmentation And For Image Segmentation Based On Prior Shape

Image segmentation is of great importance in image processing. It is a main technique of shape recovery, objects detection, recognition and tracking and has been widely used in medical diagnoses, computer-aided surgery, video based monitoring, and so on.Deformable models are very popular in image segmentation. These methods can be classified into two categories: parametric deformable models and geometric deformable models. The second one is mainly refer to level set method, which has the same formulation for the 2D and 3D problems and can overcome topology variations naturally. Variational level set methods combining the merits of variational principles and level set methods have received considerable attention recent years. These methods can integrate edges, regions, textures and motion information in image segmentation models conveniently, but there are some problems need to overcome in multiphase image segmentation and prior shape based image segmentation which is focus of this thesis. Main contributions of the paper include:1) Some variational level set methods for image segmentation appeared recent years are reviewed, and their demerits are presented also.2) A general formulation for multiphase image segmentation is proposed, which based on geodesic active contours and general probabilistie distribution region models and has a general region partitioning expression.3) A prior shape based variational level set method is proposed, which finishes registration and segmentation in a same model. The prior shape is obtained by level set method also.In addition, Constraints of signed distance functions are enforced in the models to avoid reinitializationin the classic level set methods.