Synthetic aperture sonar images segmentation using dynamical modelling analysis

Symbolic Models applied to Synthetic Aperture Sonar images are proposed in order to assess the validity and reliability of use of such models and
evaluate how effective they can be in terms of image classification and segmentation. We developed an approach for the description of sonar images where the pixels
distribution can be transformed into points in the symbolic space in a similar way as symbolic space can encode a trajectory of a dynamical system. One of the main
characteristic of approach is that points in the symbolic space are mapped respecting dynamical rules and, as a consequence, it can possible to calculate quantities that
characterize the dynamical system, such as Fractal Dimension (D), Shannon Entropy (H) and the amount of information of the image. It also showed potential to classify
image sub-patterns based on the textural characteristics of the seabed. The proposed method reached a reasonable degree of success with results compatible with the
classical techniques described in literature.