Image Segmentation Based On Active Contour Model

Image segmentation is the process that divides the whole image region into several parts. As a fundamental operation, image segmentation provides the preparative works for the high-level operations in image processing and computer vision. Recently, the segmentation algorithms based on active contours have been widely paid attention by many internal and foreign researchers due to their variable form, flexible structure and excellent performance, this paper has performed in-depth researches on the kind of segmentation algorithms. According to the representation of active contours, the content of this paper is partitioned into the following three parts:Firstly, this paper has studied the segmentation algorithms based on parametric active contours. Parametric active contours represent their contours by the point sets or B-splines, and they have the efficient performance on image segmentation. This paper has introduced the most classical parametric active contour, i.e., snake model, and its improved models, i.e., balloon snake model and GVF snake model. Through combining the good properties of balloon snake model and GVF snake model, this paper has proposed GVF-balloon snake model. The segmental experiments show that the proposed model not only preserves the GVF snake model's property of bidirectional motion, but also has the excellent performance on extracting the objects with complex shapes like the balloon snake model.Secondly, this paper has studied the segmentation algorithms based on traditional level set active contours. Traditional level set active contours represent their contours by the zero level sets of the level set functions, which is set to be signed distance functions. Traditional level set active contours have the ability of automatically handling the topology change. This paper has introduced three classical traditional level set active contours, i.e., geometric active contour, geodesic active contour and Chan-Vese model. Based on these works, we propose directional geodesic active contour and dual Chan-Vese model, respectively. The experiments conducted on image segmentation show that the two proposed models improve the performances of the original models. Moreover, this paper has applied the shape based Chan-Vese model to the extraction of circular objects, i.e., circle detection.Lastly, this paper has studied the segmentation algorithms based on binary level set active contours. Binary level set active contours represent their contours by the interface of the binary level set functions, which just take -1 and 1. Binary level set active contours not only have the ability of automatically handling the topology change, but also have high computational efficiency. This paper has introduced the region based binary level set active contour proposed by Tai et al. Aim at the shortcoming that the model proposed by Tai et al lost the gradual property of curve evolution, this paper proposed an improved region based binary level set active contour, the contour of improved model can conform to the object in the gradual evolution form, thus it preserves the motion form of the contour of traditional level set active contours. And, under the framework of geometric active contour, this paper proposed a novel binary level set active contour, which extend the application of the binary level set method from the region based image segmentation to the boundary based image segmentation.