Seafloor Texture Classification with a Multiscale Discriminant Analysis on High Resolution Sonar Images

This paper presents a robust approach for the automatic classification of sonar pictures. The classification task deals with t h e segmentation of the sea-bottom thanks t o the observations given by a high resolution sonar antenna. The originality of our approach consists in describing the seabed features with statistic multiresolution parameters. The obtained statistic parameters form a feature vector corresponding t o a scale parameter description. A discriminant analysis allow us t o significatively decrease t h e size of the vectorial space corresponding t o the feature vectors and t o generate a n optimal subspace. A training set of 300 sonar observations has been used t o reduce the feature space. This method has been validated on real world sonar pictures, with strong speckle noise.