||In conventional fractal image compression the decoding amounts to iterating an affine map on an arbitrary initial image until convergence. It has been observed that the convergence of the decoding can be accelerated by updating each pixel as soon as its new value is available. However, no analysis was provided on the dependence of this algorithm on the order in which the pixels are decoded. In this paper a technique is proposed where the ordering is based on the frequency with which a pixel was used in the fractal code. Simulations on several images show that this approach enables a faster convergence than the natural method where the pixels are decoded according to the order in which the ranges were encoded.