Image Segmentation By Clustering Algorithm

Image segmentation is a hot subject nowadays, and a broad application prospect will be got from it. It is broadly used in medicine,traffic,military and other domains.It can be used to help us to deeply understand and analyse images'information.On the other hand,it is also a very complex and difficult technique.The defects of image segmentation force researchers to innovate and improve the algorithms ceaselessly,so that the algorithms will have a good performance in the practical application.There are many algorithms about image segmentation which come from different theories.This article focus on the image segmentation based on fuzzy clustering algorithm.Image segmentation based on clustering algorithm is very important and used broadly, and many new algorithms are proposed recently. First, theories, significance and applications on image segmentation are introduced in this article.Then, theories and applications on image segmentation based on clustering algorithm are introduced so that readers will have a deeply realization about clustering algorithm.In the chapter 3 and chapter 4, three FCM algorithms are proposed. One of them aims at color image segmentation.It is extended to color image segmentation from the clustering algorithm of spatial patterns. Each pixel of color image is composed of three elements,called RGB elements.But the clustering algorithm of spatial patterns is used in the linear dimension of gray-scale value and color images'color space is three dimensions.So,a method is used here to transfer the three color values to a brightness value which is in the linear dimension.Then,it can be used in the algorithm which instead of the gray-scale value of distance formula.This proves good in the color image segmentation experiment.The others aim at noisy image segmentation. The clustering algorithms segmentation results will be affected if a image contains noisies.So,two algorithms are proposed to resist the noisies'influence.One method is used in the clustering algorithm of spatial patterns.The algorithm is modified by the membership function which is added a variable in order to anti noises as the membership matrix storages the segmentation result. So,it is meaningful in theory.The other method is used in a common fcm algorithm.The algorithm is modified by the distance formula which is also added a variable in order to resist the noises as the distance values will influence the membership values.At last,the method which is used in the former anti noises algorithm is also used in the second algorithm to intensify its anti noisies ability.It is also meaningful in theory.In noisy images experiments,these two algorithms get good segmentation results.The last chapter summarizes the paper's main contents.