Now it is widely accepted that many digital images are phase portraits of complex dynamical systems. The distribution of system trajectories in the phase space may be described by a measure that follows an exponent law. In this work we consider methods of obtaining classification signs based on the calculation of alpha-divergences (Regny divergences), the Hausdorf dimension of a measure support and averaged singularity exponents. For an given image a discrete normed measure and the sequence of measures obtained from the initial one by the direct multifractal transform are considered. In the first method to compare two images we calculate alpha-divergence between the measures from corresponding sequences. The obtained vector is a characteristic of similarity of images structures. In the second method we calculate the Hausdorf dimension of the measure support and the averaged singularity exponent. The results of numerical experiments for Brodatz textures and biomedical preparation images are given.