Gaussian Kernelized Fuzzy c-means with Spatial Information Algorithm for Image Segmentation Cuiyin Lui 1,2,3 1 College of Computer Science, Sichuan University, Chengdu 610064, China 2 State Key Laboratory of Fundamental Science on Synthetic Vision, Chengdu 610064, China 3 College of Computer Science, Panzhihua University, Sichuan Panzhihua 610007, China Email: liucuiyin@163.com Xiuqiong Zhang 1,2 , Xiaofeng Li 1,2 ,Yani Liu 4 , Jun Yang 1,2,3 1 College of Computer Science, Sichuan University, Chengdu 610064, China 2 State Key Laboratory of Fundamental Science on Synthetic Vision, Chengdu 610064, China 2 College of Computer Science, Sichuan Normal University, Chengdu, 610066, China 3 College of Computer Science, Panzhihua University, Sichuan Panzhihua610007, China 4 College of Computer Science, Panzhihua University, Sichuan Panzhihua610007, China Email: zxq_03@tom.com, lixiaofeng2009@scu.edu.cn Abstract—FCM is used for image segmentation in some applications. It is based on a specific distance norm and does not use spatial information of the image, so it has some drawbacks. Various kinds of improvements have been developed to extend the adaptability, such as BFCM, SFCM and KFCM. These methods extend FCM from two aspects, one is replacing the Euclidean norm, and the other is considering the spatial information constraints for clustering. Kernel distance can improve the robustness for multi-distribution data sets. Spatial information can help eliminate the sensitivity to noises and outliers. In this paper, Gaussian kernel-based fuzzy c-means algorithm with spatial information (KSFCM) is proposed. KSFCM is more robust and adaptive. The experiment results show that KSFCM has the better performance. Index Terms—kernel, spatial information, segmentation, FCM I. INTRODUCTION Image segmentation plays an important role in a variety of applications in high level image processing. Segmentation of structures from images is an important step in image analysis that can help in visualization, automatic feature detection, image-guided surgery, and also for registration of different images. There are various kinds of methods for image segmentation, such as gray threshold method, region growing method, watershed method, gradient method, geodesic active contour, energy method, and clustering method [1]. Clustering is a popular method used for image segmentation for its simplicity and easiness to implement. The commonly used clustering algorithms are hard c- means and the fuzzy c-means algorithm. Hard c-means clustering is based on classical set theory, and it assigns an object to one cluster or not. Fuzzy c-means (FCM) clustering reported by C.J .Dunn in [1], and proposed by Bezdek J C in [2], is an unsupervised method that has been successfully applied to image segmentation. There has been considerable interest recently in the use of fuzzy segmentation methods, which retain more information from the original image than hard segmentation method [3]. Fuzzy clustering methods allow objects to belong to several clusters simultaneously with different degrees of membership [4]. In many real applications, fuzzy clustering is more natural than hard c-clustering, in segmentation of an unclear image. It is difficult to decide a pixel belonging to which one cluster, but rather membership degrees between 0 and 1, indicating its partial memberships. The original intensity-based FCM algorithm functions well in image segmentation, however, it fails to segment images corrupted by noises, outliers, other imaging artifacts and intensity in-homogeneity. The failures are induced by two problems. The first problem is detailed in [5]. Fuzzy algorithm requires a specific distance function given, which is known that different distance function, however, it is known that different distance function yield different data structures, thus FCM clustering is not robust for multiple distribution data set. The second problem is that FCM disregards spatial information of image, thus it is sensitive to noises and outliers. There are two approaches to mitigate the impact of the two problems to improve the adaptively and robustness of FCM algorithm. On the one hand, many researchers have concentrated on extending distance functions in FCM [6- 7]. A. F. Gomez-Skarmeta and M. Delgado [8] developed a Gustafson-Kessel algorithm employing an adaptive Manuscript received March 10, 2011; revised June10, 2011; accepted September 10, 2011. This work was supported by National Natural Science Foundation of China (Grant No. 60736046), National 973 Program of China (Grant No. 2009CB320803). Corresponding author: Xiaofeng Li. JOURNAL OF COMPUTERS, VOL. 7, NO. 6, JUNE 2012 1511 © 2012 ACADEMY PUBLISHER doi:10.4304/jcp.7.6.1511-1518