3D snakes for the segmentation of buried mines in 3D acoustic images

In this paper, we describe some image processing techniques for the analysis of 3D acoustical data. More specifically, the 3D images are segmented using a deformable template (3D snake). This iterative algorithm provides a triangulated surface of the echo generated by buried underwater mines. The segmentation result can then be used for recognition/classification of the detected object purpose. The proposed algorithm consists in iteratively deforming a triangulated 3D surface until it fits the shape and boundaries of the object of interest. In the first section, we briefly review the classical technique for the segmentation and reconstruction of volumetric data (marching cube algorithm, enabling the fast extraction of a triangulated model of an object). Then, the proposed deformable model is described. Results obtained on real data sets provided by the GESMA are presented, demonstrating the interest of 3D deformable models for the analysis of 3D acoustical images.