In the underwater mine warfare context, potential threats are usually detected and classified by means of an Automatic Target Recognition (ATR) chain, especially in case of newly surveyed areas. However, if we can rely on a previously acquired sonar track, it is conceivable to directly compare such a track, said as reference, with a more recent one in order to detect seabed changes such as new objects lying on the seabed. To perform this change detection process, the very first step consists in geometrically aligning the reference and the repeated tracks. In this paper, we detail a block-matching approach using masked Fourier cross-correlation as a similarity metric, to carry out a fast elastic registration in a multi resolution framework. To improve the robustness of the algorithm, the resulting vector field is then filtered thanks to the navigation uncertainty, provided by the INS, along with an Inverse Distance Weighting estimate, to get rid of outliers.