Color Image Segmentation Algorithm Based On FCM Clustering

Image analysis and pattern recognition based on image segmentation,and image segmentation is the image processing is also a classic problem, the so-called image segmentation is based on gray scale texture,color and other characteristics on pixel similarity in accordance with certain criteria for classification. Grayscale image as color image contains more than richer information, so more and more researchers in the color imagesegmentation algorithm. Using data clustering method for color image segmentation is intuitive, easy to implement features, and can put the three color components to be considered as a whole, better segmentation. Fuzzy C means algorithm which (FCM) clustering algorithm is one of the most famous, and in the field of image segmentation has been widely used. Although the FCM in the field of image segmentation has been widely used, but still the following problems: (1) convergence to local extremum; (2) performance of the algorithm depends on the initial cluster centers; (3) shall be determined in advance the number of clusters; ( 4) Calculate the volume.Image segmentation is a large sample of data classification problem, the use of FCM for graph partitioning, optimization of each iteration to be calculated when the cluster centers and membership matrix, operation is very time-consuming, which limits the FCM algorithm in the application of image segmentation , but also makes direct use of image data cluster validity analysis is even more difficult. Thus, the calculation of a large obstacle to FCM for image segmentation, has never been solved.Based on the previous segmentation of color images on a large number of studies, analysis of the current status of color segmentation, as well as the main problem facing by the previous algorithm for the improvement and integration, design and implemented based on hierarchical subtractive clustering fast FCM clustering algorithm (SKFCM) segmentation of color images, while taking advantage of Xie-Beni index to determine the number of clusters. Experimental analysis shows that this method does not need pre-determined number of clusters, in the optimization without changing the clustering performance, the speed of fuzzy clustering can be significantly improved to achieve fast color image segmentation. At the same time on a subset of these fuzzy clustering, can significantly improve the computational speed, and then can use the cluster validity analysis of indicators to quickly identify the number of clusters. Main tasks are as follows: 1, research and analysis of color space in color image processing applications, according to various advantages and disadvantages of color space as well as specific characteristics of the division of the appropriate color space selection.2, the mainstream of today's color image segmentation algorithm to classify, analyze, summarize and summarized, focusing on the fuzzy C means clustering algorithm to study, analyze the advantages and disadvantages.3, for the FCM algorithm, the computational capacity, time-consuming and slow, this paper will be introduced to the FCM clustering layer in the subtraction. The first layer of the Subtractive Clustering data sets have a similar color np a subset of the data because the sample size is much smaller than n, will greatly enhance the computing speed of FCM. Subtractive clustering can also be the center of the cluster centers as the initial FCM, clustering performance by the center to reduce the impact of random initialization.4, for the FCM clustering algorithm, a predetermined number of drawbacks, consider the data points by calculating the tightness - separation of the Xie-Beni index, the validity of cluster analysis to get the best number of clusters can No manual intervention to achieve the requirements of automatic image segmentation.5, separated by a large number of image simulation, analysis of the proposed segmentation algorithm (SKFCM) the advantages and the problems.