Back-to-front Compositing of Contours

To maintain full flexibility in the choice of compositing operations, like local MIP or alpha-blending, a spatially consistent ordering of the projected voxels has to be maintained. Voxels with a gradient magnitude below a specific threshold do not provide a useful contribution to an image rendered using the contour shading model. Only about 25% of all voxels have a sufficiently high gradient magnitude, and are thus included into the RenderList structure, thus keeping the memory requirements at a reasonable level (For back-to-front rendering using shear/warp, three copies of the data are required, the MIP approach described before requires just one copy, as spatial ordering is not relevant).

Figure 4.23: Voxel ordering for back-to-front rendering: Voxels within each slice (=RenderListEntry) are sorted by gradient magnitude. Voxels which are mapped to 0 by $g(\vert G(P)\vert)$ can be skipped efficiently.

Within a RenderListEntry, voxels are sorted according to gradient magnitude. During rendering, only voxels which are not mapped to black due to their gradient magnitude (see figure 4.23) have to be considered. Voxels mapped to black due to the currently used $g(\vert G(P)\vert)$ are located at the end of a RenderList's voxels and can be efficiently skipped. Compared to the MIP-only ordering of voxels described earlier, significantly more voxels have to be rendered. Voxel skipping is only based on the gradient magnitude of a voxel, but not on the view-dependent property of being part of a contour.

mailto:mroz@cg.tuwien.ac.at.