Fuzzy Clustering And Particle Swarm Optimization In Image Segmentation

Image segmentation is an important component of digital image processing and it is a basic technique of image processing. The purpose of segmentation is to separate out target areas that people need from background. Cluster analysis is a common data analysis tools, it is the process to divide a data set into multiple similar data clusters. Clustering technology is widely used in image segmentation, data mining, pattern classification, medical diagnosis and machine learning. In the numerous segmentation algorithms,image segmentation based on clustering analysis is a very important and popular algorithm.Fuzzy C-Means clustering algorithm (FCM) is used widely in clustering analysis.FCM plays an important role in the fuzzy clustering theory. As an unsupervised clustering algorithm, FCM is good at convergence. However, FCM also has many shortcomings, it is easily affected by noises, depends on initial value and usually leads to local minimum, especially in the instance of large amounts of clustering objects.In order to resolve the two problems, we could improve the membership function of FCM and introduce the particle swarm algorithm.1. Clustering technology is widely used in image segmentation, but it has a lot of shortcomings. First, image segmentation based on clustering technology often split a single pixel, it does not take into account the characteristics of spatial information, therefore, the image is often segmented with a lot of noise points, that affects the result of image segmentation. Second, as the uncertainty of the image, it can not be determined which class the pixels belong to, because images have considerable ambiguity. The traditional clustering techniques are hard clustering, the membership of one pixel for a certain category is either 0 or 1, and this is not realistic. In order to resolve the two problems, FCM is proposed, the combination of FCM and spatial information technology can be a very good solution to both problems. FCM exhibits the robustness to noises, but the pixel spatial information is not considered in this algorithm, in the case of a large number of noises, FCM will be degraded. Based on FCM, an improved algorithm (Improve Possibilistic C-Means clustering, IPCM) is proposed for image segmentation by improving membership function, the new membership of the pixel is updated to the geometric mean value of its neighborhood membership. The experimental results show that the new algorithm can segment the image effectively and properly, and has good performance of resisting noises.2. FCM is good at convergence, but it usually leads to local minimum. Combined with particle swarm optimization algorithm and FCM, then add chaos technology, a new algorithm is proposed: Fast Fuzzy C-Means clustering based on Chaos Particle Swarm Optimization algorithm (CPSO_FFCM). The algorithm can search quickly the global optimal solution, and jump out of local minimum, and can get a better clustering effect. In order to avoid stagnation of particles in the iteration, the algorithm introduces the chaotic variables, and generates a chaos sequence based on the current global best position, this algorithm replaces randomly a particle of the particle swarm with the particle that has optimal-adaptive value in the chaos sequence. We use the algorithm in image segmentation, the experimental results show that the new algorithm can segment the image effectively and properly, and has the good robustness to noises and good adaptability.