Assessing the scope of the multifractal approach to textural characterization with statistical reconstructions of images

This paper is an attempt to assess more precisely the scope of the multifractal approach for textural characterization. We perform a series of numerical experiments to reconstruct black and white images from the multiscale moments of a measure defined by a reference image. The reconstructed images, obtained by simulated annealing, have multifractal statistics identical to the reference image. A comparison of the reconstructions with the related reference image reveals some of the strengths and limitations of the multifractal characterization.