Abstract:We present a novel method for cutting out 3D volumetric structures based on simple strokes that are drawn directly on volume rendered images. The freehand strokes roughly mark out objects of interest and background. Our system then automatically segments the regions of interest and refines their boundaries in the rendered image. These 2D segmentation results provide constraints for 3D segmentation in the volume dataset. The objects of interest are then efficiently cut out from the volume via a combination of our novel two-pass graph cuts algorithm and the pre-computed over-segmentation. Our method improves traditional, fully automatic segmentation by involving human users in the process, yet minimizing user input and providing timely feedback. Our experiments show that our method extracts volumetric structures precisely and efficiently while requiring little skill or effort from the user.

Volume Cutout