Object-based boundary properties

Abstract

While object-based image analysis specializes in using region features for object detection, it lacks the possibility to use border strength and local geometry, common in edge detection. We propose to enhance common object-based image representation with boundary features that measure strength and continuity. Using these we formulate strategies for merging regions in a partitioned image to identify potentially regular shapes. To illustrate the capacity of this approach, we apply the proposed concepts to CT bone segmentation.