Abstract: Aiming at the problem of inconsistency of common-view targets caused by local distortion in side-scan sonar images, this paper presents a contour-based registration framework for side-scan sonar images. Firstly, the target contour is extracted and the environmental context description of each contour point is calculated, with the χ² test statistics of the environmental context as the cost function. Then, the corresponding relationship of contour points is obtained by minimizing the cost of all matching points. Finally, the regularized thin-plate spline function is used to solve the image transformation relationship, and the contour is established. Correspondence of internal pixels. Experiments show that this method can effectively ensure the consistency of common-view targets, and has a certain reference value.

Key words: side-scan sonar image, shape context, shape matching, thin plate spline, contour