Region number determination in automatic image segmentation based on BKYY model selection criterion

In the feature space clustering approach to image segmentation, each cluster corresponds to a region in the image. Determination of the appropriate number of clusters (regions) has long been an open problem. In this paper, we apply the Baysian-Kullback Ying Yang (BKYY) Model Selection Criterian on this feature space clustering problem to give an appropriate number of regions in image segmentation. Experimental demonstrations show the results of image segmentation with this automatically determined number of regions.