Study Of Color Image Segmentation Technology Based On Fuzzy Clustering

Image segmentation plays an important role in the whole image project; it is not onlyone of the research keys in image processing, but also the basic of image analysis andunderstanding. As a common kind of image, the color image itself can carry moreplentiful visual information and its application is generalized, which decides theimportance of the research of the color image segmentation techniques.Fuzzy C-means (FCM) clustering algorithm is a classic fuzzy clustering analysisalgorithm which has more complete theory and was used more widely than the otherfuzzy clustering algorithms;it uses the iterative optimization of objective function toget the fuzzy partition of data set and has very good convergence. The fuzziness of theimage itself and human vision provides the theoretical feasibility for the method ofcolor image segmentation based on fuzzy clustering. The image segmentation usingFCM algorithm has the characteristics of simple and easy to realize, it doesn’t needartificial intervention in the clustering process, which can be applied in the automaticsegmentation application fields. That made the research of the color imagesegmentation technology based FCM algorithm has well theoretical foundation, butthere are some problems with it, such as the selection of the initial category number,the determinated of the initial centroids, large amount of calculation in clusteringprocess and post-processing.Based on the research of these problems, according to the shortage of randominitialization in traditional FCM,and for getting more accurate initializationautomatically, the paper proposes a clustering segmentation method combiningHill-climbing for color image, which can generate the initial centroids and the numberof clusters adaptively according to the three dimensional histogram of the image. Inaddition, a new post-processing strategy which combined the most frequency filterand region mergeing is introduced to effectively eliminate small spatial regions.Experiments show that the proposed segmentation algorithm achieves highcomputational speed, and its segmentation results are close to human perceptions.