Image Segmentation Method Based On Level Set And Its Test Platform

In the field of image processing, image segmentation plays an important role as a basis of image analysis and understanding as well as pattern recognition. The basic principle of image segmentation is to divide image into several parts or extract the objects of interest according to the relevant features of the image. These features usually include gray value, texture, shape and color. As a popular segmentation method, level set methods(LSMs) have the capability of elegantly handling the topology change and incorporating more multiple external constraints to realize accurate segmentation. Nowadays, LSMs have increasingly become a popular and hot research field in which a number of new algorithms are proposed every year in the recent decade. However, due to the level set method, segmentation problem is converted into energy minimization problem. That will cause some open problems, e.g., resolving partial differential equations(PDEs) usually leading heavy computational cost, flexibly segmenting multiple objects in image. In this paper, we propose a method to solve the problem, improved and enriched the image segmentation method based on level set as following.1) This paper proposes a fast image segmentation method based on the local Gauss and lattice Boltzmann method(LBM). In this method, firstly, the image features are modeled by the local Gauss model, which makes the segmentation method has better robustness to the noise. Secondly, the LBM can be used more effectively to solve the evolution equation, which can greatly reduce the computation. The experimental results show that the proposed algorithm is more efficient compared with the same method in the case of keeping the accuracy of segmentation;2) This paper also proposes a fast multi object image segmentation method based on multi region competition. The method uses the N level set functions to represent 2N regions which reduces the number of level set functions, and hereby accelerate the speed of segmentation. Additionally, a traversal algorithm is introduced to determine the label of each pixel instead of solving the optimization problem. In such way the efficiency of the segmentation is also improved greatly. The experiments show that this algorithm has better performance than the binary segmentation method, meanwhile, has better segmentation efficiency compared with the existing multiple segmentation methods;3) In order to facilitate test and comparison between the level set based image segmentation method, this paper designs and implements an online algorithm testing platform. This platform could realize the algorithm on-line operation, and the algorithm parameters can be set flexibly. Moreover, the platform has realized the visualization of algorithm flow chart and the segmentation result.In this thesis, we focus on improving the efficiency of the image segmentation method based on level set method. The thesis improves and enriches the existing level set method. The associated experimental results show that the method is effective, efficient and practical. In future, we will focus on the combination of LBM and other image models including the theoretical derivation and numerical implementation.