Distributed Parallel Image Signal Extrapolation Framework using Message Passing Interface

TL;DR

This paper presents a distributed parallel framework for image signal extrapolation that enables existing algorithms to run efficiently on clusters without modification, maintaining quality and improving scalability.

Contribution

It introduces a novel tiling algorithm for minimal overhead and demonstrates effective parallelization of complex extrapolation algorithms using MPI.

Findings

No negative impact on extrapolation quality

Good scaling behavior on compute clusters

Efficient tiling algorithm reduces processing time

Abstract

This paper introduces a framework for distributed parallel image signal extrapolation. Since high-quality image signal processing often comes along with a high computational complexity, a parallel execution is desirable. The proposed framework allows for the application of existing image signal extrapolation algorithms without the need to modify them for a parallel processing. The unaltered application of existing algorithms is achieved by dividing input images into overlapping tiles which are distributed to compute nodes via Message Passing Interface. In order to keep the computational overhead low, a novel image tiling algorithm is proposed. Using this algorithm, a nearly optimum tiling is possible at a very small processing time. For showing the efficacy of the framework, it is used for parallelizing a high-complexity extrapolation algorithm. Simulation results show that the proposed framework has no negative impact on extrapolation quality while at the same time offering good scaling behavior on compute clusters.