An Optimized Possibilistic C-means Clustering Image Segmentation Algorithm Based on Spatial Neighborhood Information

Abstract

AbstractThe spatial neighborhood information is very important for rejecting the influence of noise. But for most clustering algorithms, there is no neighborhood information contained in the membership functions. They are prone to produce bad segmentation results when the gray scale of each pixel is directly utilized as the input. In this paper, an optimized possibilistic c-means clustering image segmentation algorithm (CPCM_S) is proposed. In the process of optimization, the proposed CPCM_S algorithm optimizes the typicality values of pixels in the iterative process by utilizing the neighborhood information of each pixel in the image to correct some typicality values. The algorithm first calculates the typicality value using the CPCM algorithm, and then introduces neighborhood information to the calculation of the typicality value to further modify the typicality value, thereby facilitating the separation of the background and the target object, thus improving the ability of typicality values to characterize pixel correlation and improve performance of image segmentation. Experimental results and analysis show that the proposed algorithm can separate the target and background clearly for noisy images, especially for small target images compared with FCM algorithm and CPCM algorithm.KeywordsFuzzy clusteringImage segmentationSpatial neighborhood information