Edge surface detection with subvoxel accuracy from 3-D images

A method is described for edge surface detection with subvoxel accuracy for three-dimensional (3-D) images. Direct extension of the subpixel edge detection method for 2-D images into 3-D applications has to deal with an explosion of operational combinations that make the method impractical. To overcome such a shortcoming, this paper proposes a practical approach consisting of two major steps: (1) extraction of local edge information; and (2) estimation of global edge surface from local edge information. To extract the candidate edge points as local edge information, we make use of the connected three voxels located on the straight lines referred to as the edge estimation axes. The coordinates of candidate edge points are derived from intensity gradients of three neighboring voxels along the edge estimation axis. Then, a planar fitting is performed on a set of candidate edge points to estimate a global edge surface in 3-D. The key points of the method directly affecting the accuracy are the selection of the edge estimation axis and the weighting of the candidate edge points. The implementation reflecting these points is described.