Multifractal analysis based on discrete wavelet for texture classification: Application to medical magnetic resonance imaging

We show the relevance of multifractal analysis for some problems in image. This paper deals the characterization of brain tumor in magnetic resonance imaging. We introduce a declination of wavelet Leaders that recently been shown to provide practioners with a robust and efficient tool for the multifractal analysis of signals and images. We calculated new multiresolution parameters called average of wavelet coefficient and the log-cumulate derived from the wavelet leaders and we have solved the problem posed by the choice of interval regression that enters in the calculation of different parameters (h(q), D(q), ζ(q)). We analyze and compare our estimator and simulated image against wavelet leaders. We apply the approach developed on different cerebral images in order to distinguish between different tissues corresponding to the healthy and pathological.