A divisive hierarchical k-means based algorithm for image segmentation

In this paper we present a divisive hierarchical method for the analysis and segmentation of visual images. The proposed method is based on the use of the k-means method embedded in a recursive algorithm to obtain a clustering at each node of the hierarchy. The recursive algorithm determines automatically at each node a good estimate of the parameter k (the number of clusters in the k-means algorithm) based on relevant statistics. We have made several experiments with different kinds of images obtaining encouraging results showing that the method can be used effectively not only for automatic image segmentation but also for image analysis and, even more, data mining.