Robust global mosaic topology estimation for real-time applications

This paper proposes an iterative methodology for real-time robust mosaic topology inference. It tackles the problem of optimal feature selection (optimal sampling) for global estimation of image transformations. This is called IGLOS: iterative global optimal sampling. IGLOS is a unified framework for robust global image registration (optimum feature selection and model computation are considered within the same methodology). The major novelty is that it does not rely on random sampling procedures. Furthermore, by considering an optimal subset of the total number of correspondences, it naturally avoids trivial solution. IGLOS can cope with any motion parameterization and estimation technique. Applications to underwater linear global mosaics and topology estimation are presented.