Comparison of mean-square and absolute value distortion measures in fractal coding of still images

Fractal coding often blurs or smoothes images. In particular, edges are poorly coded due to the choice of the distortion measure. Indeed, mean square error possesses no edge preserving property. The solution proposed in this article is to code using fractals and to introduce the L1 norm or the absolute value distortion measure. The visual quality of the coded images with the L1 norm is improved. However, the computational complexity of the coder is drastically increased. It may be kept low by preprocessing using an edge detection scheme to select the pertinent measure. The reconstruction algorithm remains the same.