Toward large-area mosaicing for underwater scientific applications

Severe attenuation and backscatter of light fundamentally limits our ability to image extended underwater scenes. Generating a composite view or mosaic from multiple overlapping images is usually the most practical and flexible way around this limitation. In this paper, we look at the general constraints associated with imaging from underwater vehicles for scientific applications - low overlap, nonuniform lighting, and unstructured motion $and present a methodology for dealing with these constraints toward a solution of the problem of large-area global mosaicing. Our approach assumes that the extended scene is planar and determines the homographies for each image by estimating and compensating for radial distortion, topology estimation through feature-based pairwise image registration using a multiscale Harris interest point detector coupled with a feature descriptor based on Zernike moments, and global registration across all images based on the initial registration derived from the pairwise estimates. This approach is purely image based and does not assume that navigation data is available. We demonstrate the utility of our techniques using real data obtained using the Jason remotely operated vehicle (ROV) at an archaeological site covering hundreds of square meters.