Tensor Splats: Visualising Tensor Fields by Texture Mapped Volume Rendering

We describe a new method for visualising tensor fields using a textured mapped volume rendering approach, tensor-splatting. We use an image order method to calculate the 2D Gaussian splats or footprints of the projected 3D Gaussians at an arbitrary number of standard deviations from the centroid. These footprints are then mapped and composited front to back to the view plane by texture mapping within the framebuffer pipeline to effect a volume rendering. One of the features of tensor-splatting is that opacity transfer control, which can be used to emphasise the tensor shape, can be achieved trivially by hardware acceleration because it requires only remapping the opacity and colour of the composited texture splats. We illustrate our method on MR diffusion weighted tensor data.