Iterative graph cuts segmentation with local constraints

Most existing stroke-based graph cut image segmentation techniques use only intensity information of strokes to update intensity distributions of object and background. Accordingly, fluctuation effect may occur unexpectedly as a result of the global effect of regional term in the graph cut framework. In this note we present an iterative graph cuts-based image segmentation technique which incorporates local constraints. A new energy function with local constraints term generated following additional seed points is formulated and is minimized to obtain a globally optimal segmentation. We tested the method on cone-beam CT data of printed circuit board and comparatively found the strength of the proposed method in accuracy and controllability.