Research For Image Segmentation Based On Cluster Analysis And Level Set

Recently, image segmentation technology is one of the important subjects within image processing and analysis research filed throughout the world. Because of this importance, some scholars do a lot of extensive research and a variety of different segmentation algorithms have been proposed, but these algorithms are mostly aim to specific object, there are still no universal segmentation theories, so people are constantly exploring new segmentation algorithm and theory. In this paper, we firstlyanalyze several traditional image segmentation algorithms, and then propose several improved image segmentation algorithms. The main work is as follows.This paper firstly introducesthe traditional hierarchical clustering algorithm, and thencontrary to some shortcomings, an improved hierarchical clustering algorithm has been proposed. The basic idea is using multiple reference points to represent shape of a cluster, and then based on similar reference points to merge small clusters, while introduce the cluster validity index to evaluate the quality of the new cluster, and re-divide the low-quality new clusters, in order to avoid low-quality new cluster diffusion to the high-level.We apply it to image segmentation and obtain a good segmentation results.Then this paper introduces the traditional fuzzy clustering algorithm and density clustering algorithm in image segmentation. For some shortcomings of these algorithms, an improved weighted FCM image automatic segmentation algorithm and an improved image segmentation algorithm based on density have been proposed. The basic idea of the improved weighted FCM image automatic segmentation algorithmis:firstly, the numbers of clusters automatically getfrom the gradation-gradient histogram of an image, then use the weighted FCM algorithm, achieve the image segmentation ultimately.Then an improved image segmentation algorithm based on density is introduced. The algorithm using the triangle principles to quickly get the correlation between the pixels, greatly reduces the computational complexity, and improves the efficiency of image segmentation. The experimental result is better than traditional density clustering algorithm.At last is an image segmentation method based on improved fast implicit level set method. The proposed method is based on techniques of piecewise constant and piecewise smooth’s Chan-Vese Model, semi-implicit additive operator splitting (AOS) scheme for image segmentation. Different from traditional models, our model uses the level set which are corresponding to ordinary differential equation (ODE). Our model has more improved characteristics than traditional models, such as:less sensibility of noise; unnecessary of re-initialization and high speed by the simplified ordinary differential function.