In this communication, we present an original unsupervised image segmentation procedure which assumes the 2-D objects to be fractal. This technique is applied to the evaluation of the covering rate of algae deposit in the 'green tide' phenomenon which occurs on the coasts of Brittany. After a discussion relative to the fractal nature of the objects under study, we introduce a fractal growth model called DLA which, in conjunction with the image data, allows the obtention of a binarized image. For this, a Bayesian formulation is adopted. Some experimental results are presented, which show the potentiality of this approach.