Forward-looking sonar image registration using polar transform

This paper presents a novel approach for forward-looking sonar (FLS) image registration. By simplifying the sonar imaging motion model, we show that the transformation between two consecutive and near-consecutive sonar images approximately follows a rigid transformation. Then we propose a polar transform based approach for FLS image registration. We extract Gabor feature points and use them as the centers of the sampling regions for the polar transform. To reduce the influence of oversampling at the fovea, a Weighted Angular Projection Function (WAPF) is used to calculate the distance between two points in two images. This will determine the corresponding point pairs between the reference and target images. Then, the rotation between the two images can be calculated from the translation between the WAPFs of the corresponding point pairs. The translation between the two images is then calculated by using the Cartesian coordinates of the centers. Experimental results show that our proposed approach produced superior results compared with the state-of-the-art.