Fuzzy C-means Clustering-based Color Image Segmentation Method

As an important approach of human perception, color information is the crucial basis for pattern recognition and computer vision. Besides high dimensional color space processing is possible thanks to the cost of hardware for color image collection and processing is reduced and computer process ability is enhanced. Color image segmentation is therefore concerned. Many color image segmentation methods are directly citing the gray level image segmentation methods to three components of color separately, which possibly cause information lose and made the evaluation of result difficult. Clustering methods have the ability of dealing with high dimensional data and translating the color image pixel information to high dimensional character space which made the segmentation result more reasonable. Fuzzy set theory has good ability to describe the uncertainty and ambiguity of images, so this paper mainly research on the Fuzzy C means color image segmentation method and how to utilize the space information to make the segmentation results more robust to noises.First ideal data sets is construct to analyze the uniformization of K-means and Fuzzy C means algorithm and point out that both of the two algorithms have absolute sample distance uniformization and relative sample quantity uniformization effect, and the relationship between separate degree of clusters and clustering results is analyzed in this part. Then Iris data set and Glass data set are introduced to compare the high dimensional real data sets processing ability of K-means, Fuzzy C means and Mean shift methods, and results demonstrate that Fuzzy C means excels in dealing with real high dimensional data sets. At last Fuzzy C means color image segmentation method in which space information and color information are combined together is proposed in this paper. Segmentation results are compared with results of Mean shift color image segmentation method which shows that the algorithm proposed in this paper is effective and robust to noises and can get more reasonable results.