Statistical discrimination of seabed textures in sonar images using co-occurrence statistics

We propose a general framework to evaluate similarities from cooccurrence statistics between seabed textures within sonar images. Our approach aim at defining a texture based similarity as a weighted sum of the distances between cooccurrence distributions for a considered set of interaction types. The weighting factor introduces the discrimination power of each distribution and also cope, with the dependence of the cooccurrence statistics on the incidence angle. Classification experiments of the proposed approach show improved classification results with the weighting scheme compared to the method where the cooccurrence parameters are randomly chosen.