Efficient GPU-accelerated computation of isosurface similarity maps

We present an efficient GPU-based solution to compute isosurface similarity maps for scientific volume data sets. Our approach first replaces exact isosurface extraction with a binary volume indicating whether each voxel intersects the surface or not. We then employ bounding volume hierarchy (BVH)-trees to speed up the distance field computation. Finally, a self-similarity map is generated from which we identify representative isosurfaces. We apply our approach to compute isosurface similarity maps from different volume data sets of varying sizes and characteristics. The results demonstrate significant speed gain with acceptable loss of accuracy, showing the potential of our solution for handling large-scale time-varying multivariate data sets.