Hardware assisted volume rendering using texture mapping can provide interactive visualization of 3D scalar fields but the limited amount of texture memory is a serious constraint. For data sets exceeding the limits of physical texture memory graphics libraries often apply texture bricking but such brute-force methods severely hamper the interactivity of the rendering.
Fortunately, most data sets have large regions that don't contain interesting data. This paves the way for the application of multiresolution representations of the volume data. Such representations allow regions of interest to be rendered at higher resolutions than other parts of the data set, allowing interactive rendering of data whose uniform grid is much larger than texture memory. A second benefit is the reduction of trilinear interpolations. Since coarser levels provide a filtered, more compact, representation of the original data, resampling based upon the level-of-detail is desirable because it reduces the number of expensive trilinear interpolations. The problem with multiresolution methods is the introduction of rendering artifacts when adjoining regions differ in the level-of-detail.
We developed a method for hierarchically subdividing the data. Care is taken to insure interpolation consistency between levels while maintaining a minimal amount of data replication. Continuity at level boundaries can be provided using a specific rendering scheme, which identifies and removes the erroneously rendered regions using standard polygon clipping.
When applying slice distances adapted to the level of data representation partially overlapping slice polygons cause artifacts at boundaries of adjacent bricks rendered on different level. The artifacts can be removed by determining and clipping away the erroneously rendered regions.
