A method for automatically identifying landmark regions in 3D reconstructed CT images of vertebra is proposed using the Gaussian and mean curvature of each point. This method samples the anatomical regions in vertebra CT images which are critical in registration. These regions are well distinguished anatomical regions on a vertebra which can be used in surface registration. In addition these regions provide a road map for the surgeon to sample the required surface regions intra-operatively without the need to sample the whole surface which results in a reduction in time. These points are then registered to the points obtained intra-operatively using ICP algorithm with a new initialization method which minimizes the Hausdorff distance of the points’ curvature to find an initial match between the two sets.