Fractal encoding is computationally intensive because of the large number of domains that must be searched for each range cell and because of the computations that must be performed for each domain-range comparison. Early implementations of fractal encoding were notorious for the amount of computation time required, typically taking many hours, and sometimes days, on the most powerful UNIX workstations. This time requirement hindered the acceptance of fractal image compression as a practical method. Attempts to improve encoding speed have focused on two areas. Classification of domains can significantly speed up encoding performance by reducing the number of domains that must be searched. Most fractal image compression implementations incorporate some type of domain classification. A second approach is to reduce the number of computations required to compare domains and ranges. This can be accomplished through feature extraction. The fastest approaches combine feature extraction with domain classification search strategies.
This chapter looks at the approach first introduced in Welstead (1997), which combines feature extraction with a domain classification and search strategy based on a self-organizing neural network.
Online access to SPIE eBooks is limited to subscribing institutions.