ON THE USE OF FRACTAL CONCEPTS IN IMAGE-ANALYSIS

We discuss quantitatively the problem linked with the fractal and multifractal characterisation of complex real systems from image analysis. We have used a model of early spinodal decomposition introduced by Cahn. This model, although resulting from a purely random process, contains a characteristic length scale lambda. Using the classical box counting method, we have studied the scaling properties of this system which is by construction a non-(multi)fractal object. We show how in that case a blind use of the box counting may lead to a spurious (scale dependent) multifractal curve and how the more general scaling of the information entropy in the cross-over region around lambda can provide a method of characterisation of the morphology of the model.