While different techniques for finding dense correspondences in images taken in air have achieved significant success, application of these techniques to underwater imagery still presents a serious challenge, especially in the case of “monocular stereo” when images constituting a stereo pair are acquired asynchronously. This is generally because of the poor image quality which is inherent to imaging in aquatic environments (blurriness, range-dependent brightness and color variations, time-varying water column disturbances, etc.). The goal of this research is to develop a technique resulting in maximal number of successful matches (conjugate points) in two overlapping images. We propose a quasi-dense matching approach which works reliably for underwater imagery. The proposed approach starts with a sparse set of highly robust matches (seeds) and expands pair-wise matches into their neighborhoods. The Adaptive Least Square Matching (ALSM) is used during the search process to establish new matches to increase the robustness of the solution and avoid mismatches. Experiments on a typical underwater image dataset demonstrate promising results.
Journal or Conference Title
ASPRS 2014 Annual Conference
H. Hu, Rzhanov, Y., and Boyen, T., “Robust Quasi-dense Matching Approach for Underwater Images”, American Society for Photogrammetry and Remote Sensing (ASPRS) 2014. Louisville, KY, 2014.