Parallel implementation of 2-dimensional Toeplitz solver on MasParwith applications to image restoration

摘要

Image restoration problems can be transformed into problems ofsolving a linear system Tx=b where T is a block-Toeplitz ornear-block-Toeplitz matrix. However for many of these problems, the sizeof the matrix T is usually very large. For instance, if one is tomanipulate a 1024-by-1024 pixel image, then T will be of the order 10242-by-10242. The authors implement a parallelversion of their existing 2-dimensional Toeplitz solver in adata-parallel fashion on MasPar (DECmpp) by fully utilizing itsmassively parallel processing power. The package they developed isportable and easy-to-use. They demonstrate how to use the package torecover a satellite image which is blurred by atmospheric turbulence.The implementation details and performance results are also presented