Mapping arbitrary projections for volume rendering onto an array processor

The authors propose a new mapping technique for volume rendering on massively parallel computer. Previous mappings of volume rendering algorithms onto array processors either required large amounts of interprocessor communication or lacked the generality needed to deal with arbitrary rotations and perspective projections. The mapping developed here is based on an enhanced ray-tracking algorithm. Instead of tracing rays into volumeric data, each processor determines the samples of a local slice through which a ray intersects. The samples are collected and interpolated in such a way that optimal systolic design is guaranteed. This mapping algorithm achieves optimal pipeline design for back-to-front projections, handles arbitrary viewing angles and allows orthographic as well as perspective projections.<
>