Message-passing Parallel Algorithm for Bayesian Image Restoration

摘要

Image restoration centers this work where a Bayesian algorithm based on a simulated annealing scheme and a Markov image model is the approach chosen to solve the problem of recovering from a corrupted image. The computational cost of this algorithm is very high what justifies the need for its implementation on multiprocessor architectures. In this work we proved that it is possible to reduce the algorithm execution times by means of high performance computing techniques and that it is highly scalable. We have chosen MPI for the new parallel implementation. Portability issues were the main reasons for that choice, in order to easily execute our parallel algorithm both in a multicomputer and a network of PC's.