Concurrent radar image shape-from-shading on high-performance computers

A given serial version of the shape-from-shading algorithm was parallelized and improved to handle large images. Applied parallelization techniques, as well as the implemented load balancing and controlling methods are described. The concurrent (i.e. non-optimal parallel) version was ported to Meiko CS-2 and Intel PARAGON supercomputers as well as to a cluster of Indy workstations from SGI. Important results concerning the performance of the concurrent shape-from-shading implementation on those architectures are presented and compared to each other.