Parallel computing with block-iterative image reconstruction algorithms (Q803769)

The image reconstruction problem consists of maximizing the entropy function subject to linear equality constraints with a block structure. Proceeding from a special block-iterative algorithm based on optimality criteria, several vectorization and parallelization procedures are investigated. In particular parallelism within a block, and with independent blocks is tested. Numerical examples are taken from image reconstruction of a head section with up to 1024\(\times 1024\) variables. The results show a dramatic improvement of calculation time on a CRAY X- MP compared to a SUN-3 workstation.