Abstract
Since the birth of fractal image compression, there have been numerous researches, which aimed at speeding up the encoding step. One of the most innovative and promising approaches was by converting the range-domain block-matching problem to a nearest neighbor search problem. However, the conventional approach suffers from two drawbacks: quantization errors of the affine transform parameters and the large memory requirement. This paper presents some enhancements to the approach based on a new affine transform parametrization. Experiments showed that our new technique is able to improve the fidelity and significantly reduce memory requirement with similar encoding time and compression ratio.