Fuzzy Kernel Clustering via Reduced Set Representation with Application in Image Segmentation
Kernelized fuzzy clustering is considered more effec tive than the traditional clustering algorithms and has been widely accepted in pattern recognition community. But the directly kernelized version of FCM (fuzzy C-means) clustering usually fails to efficiently deal with tasks of pattern recognition due to the significant increase of computational complexity. To tackle this dilemma, fast schemes could be used, such as the trick used in KFCM-Ⅱ algorithm, which assumes the pre images of kernelized cluster centers could be found in the orig inal data space. Unfortunately, this assumption is not true. To reckon with the aforementioned deficiencies, this study pro poses an effective fast kernelized kernel clustering algorithm, called KFCM-Ⅲ, via reduced set representation. The pro posed algorithm uses part of data for representing the kerne lized centers and is believed more effective and efficient than its counterparts. Its validation has been demonstrated in the medical image segmentation applications. The results show KFCM-Ⅲ could outperform KFCM-Ⅰ, KFCM-Ⅱ and FCM algorithms via tuned parameterization.
Kernel tricks fuzzy kernel clustering image segmentation reduced representation
Liang Liao Ling Ouyang Dongyun Wang Xiaowei Song
School of Computer Science, Northwestern Polytechnic University, Xian, China School of Electronics School of Electronics and Information Engineering, Zhongyuan University of Technology, Zhengzhou, Ch
国际会议
成都
英文
952-957
2010-12-17（万方平台首次上网日期，不代表论文的发表时间）