A-PARM: Adaptive Division of Sub-cells in the PARM for Efficient Volume Ray Casting

Abstract

The PARM is a data structure to ensure interactive frame rates on a PC platform for CPU-based volume ray casting. After determining candidate cells that contribute to the final images, it partitions each candidate cell into several sub-cells. Then, it stores trilinearly interpolated scalar value and an index of encoded gradient vector for each sub-cell. Since the information that requires time-consuming computations is already stored in the structure, the rendering time is reduced. However, it requires huge memory space because most precomputed values are loaded in the system memory. We solve it by adaptively dividing candidate cells into different sub-cells. That is, we divide a candidate cell in which the gradient is strictly changed into a large number of sub-cells, and vice versa. By this approach, we acquire moderate images while reducing the memory size.