Depth-rectified statistical scattering modeling for deep-sea video descattering

Deep-sea exploration is widely used in fields of environment and structural monitoring as well as exploration for oil and gas, which has attracted more attention in many countries of the world. In deep-sea exploration, the scattering phenomenon seriously reduces the visual image quality. Existing methods are limited in deep-sea scattering environments with multi-depth or non-uniform illumination. Thus, a deep-sea descattering method based on a depth-rectified statistical scattering model is proposed. The model proposed uses the transmission map to model the depth-constant scattered image, and uses the Gaussian statistical model to estimate the local scattering to obtain the depth-rectified scattering map in each color channel, so as to achieve the accurate modeling of scattering at multi-depth and non-uniform illumination scenarios. In order to demonstrate the effectiveness and robustness of proposed algorithm, we conducted tests on images of different scenes in shallow sea and deep sea, as well as on video sequences in deep-sea. Experimental results show that the proposed method outperforms existing methods in subjective quality and objective evaluation.