Multiphase Image Segmentation Variational Level Set Method

Image segmentation is the basic of three dimensional reconstruction, object detection and resource classification, and it has an important application in the fields of medical diagnosis, visual monitoring, and satellite remote sensing. As the complexity of the object itself and the requirement of multi-objects to detect in images, traditional method of single level set segmentation cannot satisfy the situation. So, we need do further research work for multiphase image segmentation.Multiphase image segmentation is a new issue in this area, which main purpose is to partition the image into different areas based on different features of the image. Because of the syncretizing of many other models, the priority of topological disposal, and the representation of multi regions, varational level set methods have the priority to become a new method in image processing. In this paper, we proposed a new method using N level set functions to represent N regions, which was used in the classification of image processing, to apply in the multiphase image segmentation. The main innovation of this paper including:(1) Constraints of the level set function as signed distance function are enforced in the energy functional to avoid re-initialization to improve the calculative efficiency.(2) General regions representation model is brought out based on the region competition strategy of using N level set functions to represent N regions without overlapping and vacuum problems. Different general multiphase image segmentation models are established based on the Paragios's geodesic active region model and used in different Probability Distribution, such as, MRI image fits Gauss Probability Distribution, Ultrasound image fits Rayleigh Probability Distribution.(3) Based on the image inpainting principle, the model is extended into piecewise smooth case, and it is better to keep edge in the progress of segmentation.Finally, we validated the validity of our methods of multiphase image segmentation through the numerical experiments.