Reconstruction of distorted underwater images using robust registration

Imaging through a fluctuating air-water surface is a challenging task since light rays bent by unknown amounts lead to complex geometric distortions. This paper presents a new algorithm to undistort dynamic refractive effects. An iterative robust registration algorithm is employed to overcome the structural turbulence of the waves of the frames by registering each frame to a reference frame. To get the high-quality reference frame, the image is reconstructed by the patches selected from the sequence frames. A blind deconvolution algorithm is also performed to improve the reference frame. Experiments show our method exhibits significant improvement over other methods.