Target classification from HR sonar images

This paper presents two integrated techniques for target classification from high-resolution (HR) sonar images. Both recognition procedures start with a despeckling algorithm based on the anisotropic diffusion filter. As a second step, a fuzzymorpho-based segmentation procedure is applied to the filtered images, which partitions the image into highlights and shadow areas. A number of geometrical features are extracted from these areas, and are then used to classify targets using two techniques: (i) a Markov Chain Monte Carlo (MCMC) approach and (ii) a Decision Tree Classifier (DTC). A comparison of both recognition techniques is drawn, and classification performance is estimated by ROC curves. Very promising results are obtained.