As the depth of the octree does not reach down to cell level, the initial position of a ray might not be its exact intersection point with the visible volume. Thus, some transparent regions are still processed. We therefore introduce a cell invisibility cache to skip the remaining transparent region at cell level. We can skip the resampling and compositing in a cell if all eight samples of the cell are classified as transparent. To determine the transparency, a transfer function lookup has to be performed for each of these samples. For large zoom factors, several rays can hit the same cell and for each of these rays the same lookups would have to be performed.
A cell invisibility cache is attached at the beginning of the traditional volume raycasting pipeline. This cache is initialized in such a way that it reports every cell as visible. In other words every cell has to be classified. Now, if a ray is sent down the pipeline, every time a cell is classified invisible this information is stored in the cache. If a cell is found to be invisible, this information is stored by setting the corresponding bit in the cell invisibility cache. As the cache stores the combined information for eight samples of a cell in just one bit, this is more efficient than performing a transfer function lookup for each sample. The information stored in the cell invisibility cache remains valid as long as no transfer function modifications are performed. During the examination of the data, e.g., by changing the viewing direction, the cache fills up and the performance increases progressively.
The advantage of this technique is that no extensive computations are required when the transfer function changes. The reset of the buffer can be performed with virtually no delay, allowing fully interactive classification. As transfer function specification is a non-trivial task, minimizing delays initiated by transfer function modifications greatly increases usability.