Hierarchical Clustering Linkage for Region Merging in Interactive Image Segmentation on Dental Cone Beam Computed Tomography

Abstract

Interactive image segmentation has a better result than the automatic and manual image segmentation because a user can help the image segmentation algorithm by marking the sample of background and object in the image. The algorithm will merge the regions in the image based on the user marking. In interactive image segmentation, the calculation of the distance between regions and the sequence of the merging process is important to obtain an accurate segmentation result. In this paper, we proposed a new region merging strategy using hierarchical clustering based on interclass and intra-class variances for each region and neighborhood relationship. This research aims to improve the region merging strategy and it is expected to result better than the previous research that did not implement the hierarchical clustering. The process to segment an image concludes splitting the image into several regions, user marking to mark the sample of background and object, merging the region that is not marked by the user using the hierarchical clustering until the image fully segmented. The experimental results on dental cone beam computed tomography data show that the proposed method gives a more effective and efficient result in the segmentation process.