Performance analysis of distributed implementations of a fractal image compression algorithm

Fractal image compression provides an innovative approach to lossy image encoding, with a potential for very high compression ratios. Because of prohibitive compression times, however, the procedure has proved feasible in only a limited range of commercial applications. In this paper we demonstrate that, due to the independent nature of fractal transform encoding of individual image segments, fractal image compression performs well in a coarsegrain distributed processing system. A sequential fractal compression algorithm is optimized and parallelized to execute across distributed workstations and an SP2 parallel processor using the Parallel Virtual Machine (PVM) software. The system utilizes both static and dynamic load allocation to obtain substantial compression time speedup over the original, sequential encoding implementation. Considerations such as workload granularity and compression time versus number of processors and RMS tolerance values are also presented.