| Image segmentation is an important step from image process to image analysis, and its essence is clustering the set of the image data, so clustering analysis can be applied to the study of image segmentation. There are two key elements in the study of clustering analysis, clustering methods and estimating the best number of clusters in a set of data. The first research content which is clustering method is maturing. The second research content which is estimating the number of clusters has become hot problem from seventies. In2000, Tibshirani etc advanced the "gap statistic" method named GS method which has many properties to estimate the cluster number. Firstly, the paper studies the GS model containing gray values and the position information based on GS method, and then analysis WGS model containing multi-information which has different units of measure based on standard image data. Secondly, the paper studies k-means clustering algorithm. This algorithm selects initial points with random that may lead to the condensation points not converge and slow convergence. So this paper puts forward DSA-k-means algorithm, and DSA-k-means algorithm has been applied to WGS model, then improved algorithm has been verified by actual results through qualitative analysis and quantitative analysis. Thirdly, the dynamic weight model named DWGS model and the dynamic weight algorithm named DWGS algorithm are proposed based on improved WGS model. The algorithm is applied to image segmentation, to get the best segmentation image, the corresponding optimum number and optimal weights. Then three-dimensional image of Gap statistics, the number of clustering segmentation and weight values is analyzed and described. Finally, this paper points out the lack of the method, gives a further improvement and research directions. |
