Underwater image dehazing and denoising via curvature variation regularization

Abstract

Challenges for underwater captured image processing often lie in images degraded with haze, noise and low contrast, caused by absorption and scattering of the light during propagation. In this paper, we aim to establish a novel total variation and curvature based approach that can properly deal with these problems to achieve dehazing and denoising simultaneously. Integration with the underwater image formation model is successfully realized by formulating the global background light and the transmission map derived from the improved dark channel prior and underwater red channel prior into our variational framework respectively. Moreover, the generated non-smooth optimization problem is solved by the alternating direction method of multipliers (ADMM). Extensive experiments including real underwater image application tests and convergence curves display the significant gains of the proposed variational curvature model and developed ADMM algorithm. Qualitative and quantitative comparisons with several state-of-the-art methods as well as four evaluation metrics are further conducted to quantify the improvements of our fusion approach.