Computer Science

Graduate Research Forum Details

Network flow based algorithm for topology simplification of binary image volumes

Abstract

By closing the cortical surface at the brainstem, the human cerebral cortex is topologically equivalent to a sphere. This topological property is important for various neuroimaging applications such as human brain functional mapping. However, most image segmentation methods get the results that don't have this property. We propose a network flow based method to optimally correct the topology of a 3D solid represented as a binary image volume. Essentially, we use a greedy algorithm to correct the topology; in each stage, a non-separating cut is conducted to decrease the genus by one so that the area increase of the boundary surface is minimal. One of the novelties of this method is that we use graph cuts to compute the extent of the handles in the solid so that both geodesics and minimal surfaces can be approximated. It is also competent in time complexity when the genus number is high. Finally, we remark that the same idea also applies for topology simplification of surfaces represented as meshes and isosurfaces and have more applications in computer graphics.