Research On The Segmentation Method Based On Fuzzy Clustering

Image segmentation is the basis of image processing and is also important in the field ofcomputer vision. The quality of the results of Image segmentation will directly affect thequality of the subsequent image processing applications. Due to the image’s uncertainty andimprecision, fuzzy theory is a good way to analyze the Image segmentation. The fuzzyC-means (FCM) clustering algorithm is the most widely used cluster analysis algorithm. FCMalgorithm is applied to image segmentation and a lot of improvements are given by manyscholars. This paper will combine the incremental clustering algorithm with the FCMalgorithm, and propose a new segmentation algorithm. This paper includes the followingworks:Firstly, for the need of incremental clustering algorithm, in order to accelerate theconvergence speed and improve the noise immunity, this paper selects the FCM algorithmbased on the spatial information to improve FCM segmentation algorithm. This paper gives aimproved FCM algorithm, and the algorithm considers neighborhood of each pixel, and eachpixel is expanded to five-dimensional vector form from the single gray value, including thepixel gray value, the mean of the neighborhood, the neighborhood variance, B values andthe entropy. The experimental results show that the improved algorithm is more efficient, andhas high segmentation accuracy and strong noise immunity. Standards-based Fuzzy C-Means(FCM) clustering algorithm for image segmentation, when the sample data is large, clusteringneeds to consume a very long time. Since the standard FCM image segmentation algorithmusing only the image pixel gray level information, but does not consider the interactionbetween the pixels. Therefore, the algorithm is sensitive to noise, split prone to serious errors.FCM algorithm itself due to sensitivity to initial values, easy to fall into local minimum, sothe initial setup FCM has been an issue of concern. While traditional fuzzy clusteringalgorithm are static or offline, the image processing is performed on the whole a pictureprocessing, using incremental clustering method can enhance the dynamic and real-timeimage processing. Incremental clustering algorithms are hard clustering, segmentation highefficiency, combined with FCM algorithm by using incremental clustering results to initializeFCM, which in turn reduces the possibility of FCM algorithm into local minima.Experimental results show that the higher the efficiency of this improved algorithm for segmentation, segmentation and a high accuracy and strong anti-noise.To the FCM clustering algorithm combining with progressive paper improvedstandards-based FCM segmentation algorithm, each pixel will be expanded to five-dimensional vector form, such changes into account the neighborhood information of theimage to improve the FCM algorithm split efficiency, accuracy and segmentation noiseimmunity. To further enhance the ability of anti-noise reference FCM algorithm based onspatial information, the objective function space penalty term added to further improve thealgorithm of noise immunity, but the border is a sensitive issue facing the algorithm. Thispaper combined the improved FCM algorithm with the incremental clustering algorithm. Bythe way of setting processing threshold, the algorithm alternates the incremental clusteringand FCM clustering until finished processing all image pixels. Such a combination makes theway of data processing from the whole for once to dynamic. Also the incremental clusteringprovides the initial cluster centers and partition matrix for FCM clustering to accelerate theconvergence speed of FCM iterations, and also reduces the possibility of FCM iteration into alocal minimum. The experimental results show that the proposed incremental segmentationalgorithm based on FCM algorithm has strong noise immunity, good segmentation accuracyand high efficiency.