Region of interest on SONAR image for non symbolic registration

The registration of SONAR images, in a first step, requires to determine landmarks in the environment. We propose to apply saliency map as a solution to detect landmarks and to compare results with the commonly used corner detector of Harris. This study constitutes an initialization to a previous work on non symbolic methods to register SONAR images in the framework of an autonomous underwater vehicle's (AUV) navigation (Chailloux and Zerr, 2005). Indeed, these registration methods based on pixels intensites, or their statistical distributions, are applied in a neighborhood of these interest points. Saliency maps present the advantage, comparing with the Harris corner detector, to entirely perceive the environment of the pixel, in considering many directions of observation from a pixel. Saliency map returns better scattered interest points, and captures sharp features; finally, the spatial complementarity of both methods is used in order to spread the survey zone, and improve the estimation of the displacement vector field.