Image Segmentation Based On Spatial Fuzzy C-means

Image segmentation is the basis of image processing and low-level machine vision. According to certain characteristics, image segmentation divides the image into several non-overlapping regions, with homogenous properties in the same area and inhomogenous properties in different regions. FCM is an effective image segmentation method, it divided the data of similarity into the same class, and the data in different classes is of minimal similarity. However, FCM segments the images only depends on the pixel itself and ignores the spatial information between the pixels of the image. That makes the performance of FCM image segmentation method unideal in some cases. To overcome this problem, this paper made the following work:Firstly, chap two presents a novel spatial FCM based on Minkowski distance for image segmentation(Minkowski-s FCM). Minkowski-s FCM takes the original image pixels’ gray value and filtering image pixels’ gray value as the clustering characteristics, uses the Minkowski distance as the similarity measurement. The proposed Minkowski-s FCM approach considered the spatial information in the process of image segmentation and improved the segmentation performance of FCM algorithm. At the same time, Minkowski distance, using p norm instead of 2 norm, makes the proposed algorithm not only applies to sample of the distribution of the spherical, can also be applied to sample of other shape of the distribution. Thus the applicability of the proposed algorithm is improved.Secondly, because the traditional FCM algorithm is based on the Type 1 fuzzy set, the processing ability to uncertainty is insufficient. Aiming at this problem, this paper studies the FCM image segmentation method based on type Ⅱ fuzzy set(T2FCM). In order to improve the performance of the algorithm, two novel spatial T2FCM(s IT2 FCM 、2DIT2FCM) are proposed, considering the spatial information into the image segmentation process of T2 FCM image segmentation process.Thirdly, contraposing the drawback that the two-dimensional FCM(2DFCM) segmentation algorithm using average gray values of neighborhood pixels sothat can not effectively reflect the the neighborhood pixels’ influence degree on the center pixel, This paper proposed one weighted spatial information calculation model, using fuzzy clustering method calculating the weight of neighborhood pixels’ influence on center pixel. Thus more reasonable spatial information is obtained. Based on the weighted spatial information, a new spatial weighted two-dimensional FCM image segmentation method(w-2DFCM) is presented, further improved the performance of FCM segmentation. The competitive performance of w-2DFCM is demonstrated through experiments on different images and comparisons with other FCM image segmentation methods.