Adaptive region growing based on boundary measures

Abstract

Weak boundary contrast, inhomogeneous background and overlapped intensity distributions of the object and background are main causes that may lead to failure of boundary detection for many image segmentation methods. An adaptive region growing method based on multiple boundary measures is presented. It consists of region expansion and boundary selection processes. During the region expansion process the region grows from a seed point. The background points adjacent to the current region are examined with local boundary measures. The region is expanded by iteratively growing the most qualified points. In the boundary selection process, the object boundary is determined with the global boundary measure that evaluates the boundary completeness. Experimental results demonstrate that our algorithm is robust against weak boundary contrast, inhomogeneous background and overlapped intensity distributions. © 2011 Springer-Verlag.