Skip to Main Content
We propose a region-based segmentation of textured sonar images within a level set framework. We state image segmentation as the minimization of an energy involving region regularity constraints and a texture similarity measure adapted to sonar images (introduced in our previous work [Karoui, I., et al., 2005]). In this framework, sonar textures are characterized by statistics of their responses to a set of filters, and the similarity between texture samples is measured according to a weighted sum of the Kullback-Leibler divergence between the compared texture statistics. The texture similarity measure weights setting is twofold: first we weight each filter, according to its discrimination power, the computation of these weights are issued from the margin maximization criterion. Second, we add an additional weighting, evaluated as an angular distance between the incidence angles of the compared texture samples, to cope with the problem related to the sonar image acquisition process that lead to a variability of the backscattered (BS) value and the texture aspect with the incidence angle range. We have tested the method, using different filter response first and second order distributions, on several sonar images. The results prove the relevance of the proposed method.