The Improved Fuzzy C-Means Clustering For Noisy Image Segmentation

The research about image segmentation has attached greatly importance to people from the 1970'. It is hotspot and focus of the image technique, which has been applied for all domains about image processing. Image segmentation is usually for the sake of image analysis, recognition, tracking, understanding, compress coding etc. The veracity of segmentation is very important to the succeeding task.Fuzzy C-means (FCM) clustering segmentation is a datum clustering method based on the optimization of objective function, which partitions the special vector data point into c interspaces. The clustering result is that a datum is represented with the membership to the clustering center. However, the FCM clustering algorithm is of some flaws. On one the hand, it is very weak to restrain the noise, because the standard FCM arithmetic only used the gray feature to segment image, taking no into account the special feature, so the segmentation is not intact. On the other hand, FCM algorithm is over-segmentation easily. This paper used the gray and special feature of image pixels to segment image, which involved in two aspects:Firstly, I propose an improved FCM algorithm based on the fast FCM clustering, which uses the degree of gray similarity and cluster distribution statistics of the neighbor pixels to form a new membership function. Not only it is effective to constrain the noise, but also it is ease to correct the misclassified pixels. Experimental results on three types of noisy images indicate that the segmentations are more accurate and robust than the standard FCM algorithm.Secondly, the low contrast noise image is be segmented, which utilizes the theory of rough sets to carry on a reasonable classification. Not only it is effective to filter the noise of image, but also it is better to preserve the tiny edge details than other methods, which is very important to the fuzzy enhancement of the local contrast image. The local contrast image can be improved properly by enhancing the pixels lying in both sides of the edge. There has an obvious improvement about visual effect. Then segment the enhanced image with the fast FCM algorithm and peer group technique method. The segmentation result is better than the standard algorithm, which has no over-segmentation.