Image Dehazing Based on Second-Order Variational Model

The algorithm based on dark channel prior theory can effectively dehaze under different scenes, but the image usually contains noise and some details which are not kept effectively after dehazing. The second-order variational model takes the second-order derivative as regular term, and can be used for image denoising. It has a good edge retention effect. In this paper, first of all, the dark channel prior method is used to estimate the transmission rate of hazy images, and then it is combined with second-order variational models including Laplacian variation model, Hessian matrix variation model, total generalized variation model and total curvature variation model, respectively. Four new second-order dehazing models, namely, H-LV model, H-HMV model, H-TGV model and H-TCV model, are proposed. In order to improve the computational efficiency of proposed models, corresponding ADMM(alter direction method of multipliers) algorithms are designed. By introducing auxiliary variables, the lagrangian multiplier is continuously updated and iterated until the energy equation converges. The experimental results using LIVE Image Defogging database show that the edge of images obtained by proposed models are good and image noise can be suppressed.