Research On Algorithms For Image Segmentation Based On Fuzzy Clustering

Image segmentation is a technology that it is just to divide some image intodifferent small-images(according to some criteria) with different characters and extractthe interesting regions, Image segmentation as a key issue in the field of imagetreatment and computer visual, the results of it are very important to the image analysis,image recognition and image identification. In the numerous image segmentationalgorithms, image segmentation based on fuzzy clustering method is becoming moreand more popular. And the algorithm of fuzzy C-means (FCM) clustering algorithmis one of the most widely used in image segmentation methodsHowever, the fuzzy C-means (FCM) clustering algorithm has its inherentdrawbacks in image segmentation. For example, the fuzzy C-means (FCM) clusteringalgorithm is the processing to each pixel of the image, a great deal of time will beconsumed; the fuzzy C-means (FCM) clustering algorithm is vulnerable to the impactof the initial cluster centers, and it may converge to local minimum value; the fuzzyC-means (FCM) clustering algorithm only uses grey feature without consideration ofthe special features of pixels. Thus these shortcomings affect the results of the imagesegmentation. So image segmentation based on watershed and improved fuzzyC-means (FCM) clustering algorithm is proposed in this paper. Dividing image withthe help of watershed algorithm, and the primary results are gained. It makes full use ofthe ability of global optimization PSO (particle swarm optimization) to obtain theaccurate original cluster centers of FCM. A novel distance has been established whichcontains region’s area and region’s variance. The experimental results show that thismethod has higher segmentation speed and better segmentation results compared withthe fuzzy C-means (FCM) clustering algorithm, and it has stronger anti-noiseproperty.The number of clusters is also affecting the results of image segmentation, imagesegmentation based on auto-adaptive and improved Fuzzy C-means (FCM)Clustering is also proposed in this paper. Dividing image with the help ofmarked-watershed algorithm, and the primary results are gained, the number of smallareas is an upper limit; in order to determine the number of clusters, a cluster validityindex is introduced; It makes use of the maximum and minimum distance algorithm toobtain the accurate original cluster centers of Fuzzy C-means (FCM). A novel constraint has been established which contains region’s area and region’s variance, thisimproved Fuzzy C-means (FCM) is used to merge similar regions which are gainedby watershed algorithm. The experimental results show that this method realizessignificant the number of clusters.