Parallel implementation of an adaptive and iterative noise filter for efficient signal restoration

In signal restoration, especially image restoration, there are tremendous demands for large volume and high speed computations. The paper describes application of a novel noise filter, the adaptive differential iterative RMS filter (ADIRMS), to restoration of noisy signals without any a priori information regarding the signal. An efficient parallel implementation of the filter on distributed memory MIMD machines is outlined. Since the local memory in each processing element (PE) of the system is accessible to other PEs only via explicit message passing, the manner of data distribution and communication is a very important factor for efficient solution of a given problem; this issue is considered in some detail. The ADIRMS algorithm has been implemented on the Fujitsu AP1000 with high efficiency. We conclude that this algorithm provides a good basis for parallel implementation on distributed memory machines