Rough Set Based Fuzzy Clustering And Application In Image Segmentation

The process that a physical or abstract set of objects is partitioned into several similar objective clusters is called clustering analysis. Objects in the same subgroup are similar while they are dissimilar each other if they are in different subgroups. Here the subgroup is a mathematical word which is used as the set of objects. For the last dozens of years clustering analysis has been developed maturely. It becomes a very hot research subject in the field of data mining, including image segmentation, and so on. To meet the requirements for application, a lot of clustering methods are put forward. However, there are kinds of problems in view of different requirements for different clustering applications. So within the framework of rough-set theory, this paper comes up with a generalized fuzzy clustering method based on rough-set and genetic algorithm (GA) to cope with the problems in clustering tasks using the idea of optimizing an objective function via evolution. In addition, this clustering method is also employed in the field of image segmentation. The main contributions can be listed as follows:An improved generalized fuzzy c-means clustering method based on rough-set and GA is put forward. Within the framework of rough-set theory, this method introduces the idea of GA and designs a special chromosome encoding scheme to adopt the gene manipulation, with selection, crossover and mutation included. At the same time this method combines two different types of traditional clustering algorithms FCM (Fuzzy C-Means) and PCM (Possibilistic C-Means) in order to cluster the objects, which helps to lower the sensitiveness to initially selected cluster centers and improve the performance and stability of clustering results.According to the requirements of image processing tasks, the clustering method mentioned above is applied to the field of image segmentation, in order to achieve the goal of object identification. For a given image,.the mathematical features are extracted with the methods of wavelet decomposition and GLCM. In addition, a pre-segmentation process called watershed algorithm is introduced here to lower the redundancy and scale of objects. The result of image segmentation can be obtained after a clustering process with the method put forward in this paper. Some experiments prove that this method can strengthen the stability and improve the average accuracy.To cope with complicated objects, especially for manifold data objects, based on rough-set theory we carry out clustering with a density-sensitive manifold distance measure employed in the data space to process the objects. Hence the similarity between each object pair can be obtained. Afterwards the proposed GARFPCM clustering algorithm in this paper is used to partition the data of the subsequent eigenvector matrix. Thus we can reach the goal of classifying and recognizing manifold or complicated objects. This clustering method mentioned above improves the average accuracy and robustness in clustering manifold or complicated data.