Fuzzy Clustering Algorithm Based On Intra-class Distance And Inter-class Distance And Its Application In Image Segmentation

Image segmentation is an important technology of image, its essence is the pixel clustering problem, it is in different areas of the image according to the special meaning of division, and different regions is mutually related, each specific region with consistency. With the pattern analysis and machine intelligence technology unceasing development, image segmentation caused widespread concern and research of many scholars, which has been successfully applied in pattern recognition, machine learning, the more remote sensing observations, biomedical and other areas. Because of different image sources and complex structures, segmentation models and methods is not universal, for different types of image segmentation method is proposed, in order to obtain satisfactory segmentation results. Clustering technology is currently widely used method of image segmentation, and has some scholars attach great importance to. In many clustering techniques, fuzzy c-means clustering is a kind of using iterative method to achieve classification of samples of fast and memory space consumption less important algorithm, has become for solving the image segmentation problem is more practical segmentation technique. However, due to the clustering segmentation algorithm without considering the correlation between pixel and its neighbors, is not conducive to remote sensing and medical image segmentation requires complex spatial distribution. Later, scholars have put forward a number of other fuzzy clustering segmentation algorithm, but the segmentation algorithm mostly the distance from the sample to the cluster center which is intra class distance as a measure of the algorithm, but the segmentation algorithms only consider the within class and between the distance, without considering the class of distance, makes for some image classification is not accurate enough, and the noise is larger. Therefore, this paper studies the fuzzy clustering algorithm and its application in image segmentation based on class distance. The main work is as follows:1. Aiming at the defect of fuzzy C-means clustering algorithm based on the intra-class distance between the sample and its cluster center, a new kind of fuzzy C-means clustering algorithm is proposed and applied to image segmentation by combining the inter-class distance between clustering centers. The difference of intra-class distance and inter-class distance within the object function to be the clustering measure in the algorithm, which not only considering intra-class compactness but also taking into account the inter-class separation, and the compactness of clusters and separation within them can achieve optimal values by adjusting the relevant parameters so that improve the accuracy and robustness of image segmentation. The results of the segmentation tests of a large number of synthetic images and real remote sensing images show that the proposed improved clustering algorithm with intra-class distance and inter-class distance is effective, especially for noisy images, the effect is better than other fuzzy clustering algorithms.2. In order to further improve the noise immunity and the accuracy of image segmentation, an improved possibilistic clustering algorithm combinated intra-class distance and inter-class distance is proposed and applied to image segmentation. The constraint that memberships of sample points across clusters must sum to 1 in fuzzy c-means is removed by using the possibility measure so that its membership degree is suitable for the characterization of "typical" and "compatibility", and the inter-class distance in the objective function is introduced to optimize the separation of clusters and compactness within them simultaneously so that the image segmentation will have strong stability and better noise immunity to different clustering structure. Through the synthetic and remote sensing images segmentation tests show that the proposed improved possibilistic clustering algorithm is effective, comparing with the existing other clustering algorithms can obtain more satisfactory result of segmentation.3. Aiming at the existing nuclear spatial fuzzy c-means clustering that the kernel distance between a sample and its cluster center is only considered, a new kind of kernel space fuzzy c-means clustering algorithm mixed the intra-class and inter-class nuclear distance is raised and applied to image segmentation. The algorithm combines intra-class and inter-class distance as the measurement of the sample clustering and introduces the kernel space to gain new fuzzy clustering objective function, the corresponding membership degree and cluster center expression is obtained by using the Lagrange multiplier method to acquire the clustering objective function, a fast algorithm for image segmentation is presented by incorporating histogram clustering segmentation.. Experiments show that the proposed kernel space fuzzy clustering segmentation algorithm based on intra-class and inter-class distance has strong robustness and accuracy for the image segmentation, compared to other clustering algorithms can obtain satisfactory segmentation results.