Region Merging Strategy Using Statistical Analysis for Interactive Image Segmentation on Dental Panoramic Radiographs

Abstract

In low contrast images such as dental panoramic radiographs, the optimum parameters for automatic image segmentation is not easily determined. Semi-automatic image segmentation which is interactively guided by user is one alternative that could provide a good segmentation results. In this paper we proposed a novel strategy of region merging in interactive image segmentation using discriminant analysis on dental panoramic radiographs. A new similarity measurement among regions is introduced. This measurement merges regions which have minimal inter-class variance either with object or background cluster. Since the representative sample regions are selected by user, the similarity between merged regions with the corresponded samples could be preserved. Experimental results show that the proposed region merging strategy give a high segmentation accuracy both for low contrast and natural images.