Keywords : segmentation


Sonar images depict at the same time seafloor reflectivity (backscattered acoustic energy is more or less important depending on the seafloor type) and interface micro-relief. One very often detects textured areas with signatures carrying information: rocks areas, ripples areas, sand areas .... To interpret an image the geologist makes a "manual segmentation" of the sonar image by using complementary data such as sediment grabbed samples, sub-bottom profilers data... In order to help him for this first processing step, it is possible to compute an automated segmentation of acoustically similar acoustic areas. We present here some results issued from Imen Karoui's thesis (ENST-Bretagne). The results of this work are not yet implemented within SonarScope.

Taking into account the angular dependance

Algorithms developed by Imen Karoui take in account angular dependence of both the backscattered intensity and the texture.

The backscattered energy level varies according to the incidence angle. The "raw" image given here directly comes from the sonar. One generally prefers to apply a series of pre-processing to calibrate the image, however the segmentation process presented here works quite well on this type of image.

This image is acquired on a uniform sand waves field; according to the incidence angle the sonar provides quite different textures.