Seafloor characterization by bathymetric image segmentation

The aim of this work is the characterization of seafloor using bathymetric images segmentation. The bathy-metric image is considered as set of oriented wave profiles generated, in general, by unidirectional currents (water flows constantly in one direction) where each profile can be viewed as the superposition of fast oscillations (ripples) and slow oscillations (sand waves). Due to the texture and multicomponent aspects of the bathymetric image, the characterization is performed using an adaptive mutliresolution decomposition based on the multivariate empirical mode decomposition (MEMD) technique. The MEMD a data driven technique, well dedicated for non-stationary and multicomponent data analysis, is applied to multivariate bathymetric profiles to generate 2D intrinsic empirical functions (IMFs) or empirical images each one associated to a bedform such as ripple, sand wave or dune (migaripple). Preliminary results on real bathymetric images shows the interest of the MEMD as a feasible decomposition tool to identify main bedforms.