Non symbolic methods to register SONAR images

To register side-scan SONAR images we have adapted some non-symbolic methods, using similarity measures based on phase correlation, mutual information or correlation ratio. The purpose is to correct the Autonomous Underwater Vehicle´s (AUV) position by compensating the spatial drift induced by the Inertial Navigation System (INS). These methods, based on pixels intensites, or their statistical distributions, provide many advantages; (1) avoid a costly time process (in terms of segmentation); this allows real time computing; (2) use non-supervised methods. (3) allow an adaptability to a varying seabed, often including distortions, illumination variations, occultations, or simply noise contribution; (4) increase the precision of localization. Each registration operates on an overlapping zone, between a geo-referenced image and a test one acquired along the AUV course. The registration result is a vector field, allowing the correction of the localization of the AUV, in real time. Applied on two data image sets, these methods reveal their performances following the lie of the seabed. Comparisons and possible complementarities of these methods will be shown later.