Anti-Aliased Volume Extraction

We present a technique to extract regions from a volumetric dataset without introducing any aliasing so that the extracted volume can be explored using direct volume rendering techniques. Extracting regions using binary masks generated by contemporary segmentation approaches typically introduces aliasing at the boundary of the extracted regions. This aliasing is especially visible when the dataset is visualized using direct volume rendering. Our algorithm uses the binary mask only to locate the boundary. The main idea of the algorithm is to retain the natural fuzziness at the boundary of a region even after it is extracted. To achieve that, intensities of the boundary voxels are flipped so that they are now representing a fuzzy boundary with the empty region surrounding it, while preserving the boundary position.