Study On Image Segmentation Based On Spatial Information Constraint Clustering Algorithm

Image segmentation is one of key tasks in image processing and computer vision. Many clustering-based methods have been proposed for image segmentation. Among these clustering methods, one of the most popular methods for image segmentation is Fuzzy C-Means (FCM) algorithm. However, the traditional FCM is very sensitive to noises, outliers and other image artifacts because all of pixels in the image are regarded as separated points without any connections. In order to overcome the disadvantage of FCM algorithm, this paper proposes an extended fuzzy local information c-means clustering algorithm (EFLICM) and a weighted local information based fuzzy c-means clustering algorithm for image segmentation (LSFCM) respectively. These algorithms improve the robustness to noise, especially for images corrupted by intense noise. The contributions are:1) The local distance information with membership degrees:In this paper, we propose a weighted local distance term which contains local distance information with membership degrees of neighborhood pixels. EFLICM algorithm and LSFCM algorithm both used this method. By this method, the weighting local distance information is able to exactly reflect the distance information of the same class and the different class so that the non-noisy pixels in the neighborhood enhance the influence on the center pixel while the influence of the noisy pixel in the neighborhoods suppressed.2) The weighted local gray term:In fact, if a neighborhood pixel is corrupted by the noise, it should have less contribution to the central pixel. In contrast, if a neighborhood pixel is similar to the central pixel, it may have more contributions to the central pixel. So, a weighted local gray term in LSFCM algorithm is proposed to measure the similarity between the pixel and its neighbors.3) Free of using any parameter:To improve the performance of FCM algorithm, many modified FCM algorithms have been proposed, such as FCMS, EnFCM, FGFCM, NWFCM. There is a crucial parameter used in their objective functions to balance between robustness to noise and effectiveness of preserving the details of the image. Generally, its selection has to be made by experience or trial and error experiments. The parameter in EFLICM algorithm is 0.05, and LSFCM algorithm is completely free of using any parameter that controls the balance between the image noise and the image details. Besides, the FCM algorithm and its variants make some achievements for the image without noise or the image with low noise. When the noise increasing, the segmentation accuracies of all of them decline dramatically even disabled. In this paper, we proposed two clustering algorithm which are not only suitable for low noise, but also good at dealing with images with intense noise.Experiments implemented on synthetic and real-world images demonstrate that the proposed EFLICM algorithm and LSFCM algorithm achieve better performance for image segmentation, especially for images corrupted by intense noise, compared to the traditional FCM and its extended methods. About these two algorithms, LSFCM algorithm is more stable and keeps more image information about fine structures or textures than EFLICM algorithm.