Kernel-based Fuzzy Clustering Algorithm Research

Image segmentation refers to divide images into a series of non-overlapped homogeneous regions. It is the basic problem of image processing and computer vision, and the key step from image processing to image analysis, and then to complete image understanding. Image segmentation has been widely applied in areas such as computer vision, pattern recognition and medical image processing.Among the plenty of image segmentation algorithms, Fuzzy C-Means algorithm is widely used. The objective function of traditional Fuzzy C-Means clustering algorithm is created by adapting Euclidean distance, and the classification results by using the algorithm have some shortcomings. At present, many researchers improve the traditional fuzzy clustering algorithm by using different way to define the distance of objective function. Howerve, the way of using kernel distance of Kernel Function, which is used to build objective function, to design Kernel-based Fuzzy clustering algorithm is widely applied.Although Kernel-based Fuzzy clustering algorithm improves the classification ability of the algorithm, but the algorithm ignores the spatial continuity of image information, and is too sensitive to the noise and lack of robustness. In addition, the accuracy of calculating the cluster center is not high while the algorithm deals with the sample set contain noise. In view of the above problems, this paper makes the further discussion and research in the Kernel-based Fuzzy clustering algorithm, use Noise Class and Markov random to the algorithm to enhance the segmentation capacity. The main research contents are as follows:1. Introduce the relevant content of Fuzzy C-Means clustering algorithm and explain the advantages and disadvantages of the algorithm. On this basis, introduce the definition and the basic principles associate with Kernel Function as well as the choice of Kernel Function, analyze the idea of Kernel-based Fuzzy clustering algorithm and verify the algorithms increase the effectiveness of traditional Fuzzy C-means segmentation algorithm thorough the simulation experiment.2. In order to improve clustering capability of the algorithm, the paper integrates the concept of Noise Class into the Kernel-based Fuzzy clustering algorithm and designs a Fuzzy Kernel clustering algorithm based on Noise Class. Finally, verifies the validity of the algorithm through the simulation experiments.3. For the purpose of further improve the algorithm accuracy, the paper designs the Fuzzy Kernel clustering algorithm based on Markov Random Field and Noise Class. In the way of the algorithm design, integrate a new prior mode with Markov Random Field, which is based on the membership matrix, and introduces the concept of Refuse Accept Degree to the algorithm. Finally, apply the algorithm to the artificial texture images and medical images to verify the feasibility and superiority of the algorithm.The presented algorithms in this paper are all realized in MATLAB environment, compare the experimental results with other correlation algorithms and are verified the feasible and effective.