Impact of parallel computing on fractal image compression

Abstract

Fractal theory is emerging as a dominant force in the area of image compression. The resulting images are strikingly good even at very high compression rates and the technique additionally shows promise for simultaneous rectification of the image. Although fractal image decompression is a relatively inexpensive operation, widespread use of the technology is limited by the computational complexity of the fractal compression itself. This complexity results from a search for contractive regions of self-similarity within the image. Conceptually the problem is that of irregular parallelism. The search can begin in a highly parallel fashion but must become increasingly dependent as the process converges. In this paper, we employ macro parallelism techniques based upon loosely communicating computers each assigned its own region of the search space. As the search progresses, each computer periodically broadcasts its search status and all cooperatively readjust to speed the process. The techniques advanced are expected to have wide utility since the approach utilizes nets of readily available computers rather than either supercomputer or dedicated silicon.