Table 1: Comparison of segmentation time and number of iterations for different implementations of SCAPE. The parallel hardware (GPGPU) implementation has faster computation times than the single processor implementations of both FBA and Dijkstra’s algorithm. Altogether, the improvements to the CPU implementation of SCAPE reduce the segmentation time by 90%.

CPUCPU
no improve all improveDijkstra GPGPU

Image size time (s) iter. time (s) iter. time (s) time (s) iter.

256 × 256 × 124 334 1229 28 219 12 2.4 1229
512 × 512 × 426 4975 1241 189 124 * 37 1241

*Dijkstra implementation failed due to memory constraints. The test system had 12 GB of RAM.