Research Of Algorithm For Image Segmentation Based On The C-means Clustering

Image segmentation is the technology and process which extracts the interested target, itis the foundation of image analysis and understanding which can divide the image intodifferent areas by different characteristics. Image segmentation occupies an important positionin the image engineering.The fuzzy theory has a better description to the existing uncertainty of he image, fuzzyclustering can reflect the reality of the world through the description of the intermediary ofthe sample category, so fuzzy theory gradually become the mainstream of the cluster analysis,fuzzy clustering is widely used in image segmentation and obtained good results. The fuzzyc-means clustering algorithm is one of the most sensitive and most widely used algorithm inmany fuzzy clustering algorithm, it is widely used in the application of image segmentation.But this algorithm are particularly sensitive to the initial values and it’s resistance to the noiseis not good, so it is very easy to fall into local minimum and can’t get the global optimalsolution; Moreover, this algorithm must know the clustering numbers of categories, but theclustering numbers is generally difficult to know in advance. Through the understanding andanalysis about the operation mechanism of the fuzzy c-means clustering algorithm, this paperimprove the algorithm from two aspects.First of all, we select the clustering number of segmentation through the statistical ofhistogram and reduce the dependence of algorithm to knowledge and experience. we make theimage clear clearer through the histogram equalization and then the gray range of the imagebecome bigger; The adaptive filtering processing keep the boundary and the high frequencypart of the image, it produces better effect than linear filtering and reduce the influence ofnoise to the segmentation results; Making full use of neighborhood gray information of theimage instead of traditional grey information, we can effectively remove noise to thesegmentation effect; Experimental results show that this algorithm can reduce the noise betterand improve the segmentation accuracy of the algorithm.Secondly, due to the standard FCM algorithm just use the gray information of image,segmentation is problem. This paper defined clustering category number roughly by using theneighborhood relationship attribute of the two-dimensional histogram of image, weightedeuclidean distance instead traditional euclidean distance and improve the objective function of clustering.The experimental results show that the improved algorithm have bettersegmentation effect and improve the algorithm’s robustness, the advantages is particularlyobvious in processing medical image with noise. Improved FCM algorithm can effectivelyovercome the sensitive problem to initial value of the traditional FCM algorithm and improvethe accuracy of segmentation.