The atmospheric light scattering and absorption from the aerosols usually cause visible light imaging quality decline. The dark channel prior method is a dehazing method to estimate transmittance or background light, which is suitable for foggy image restoration with the influence of suspended particles. Based on the DCP model, the proposed method combines mist concentration level estimation and nonlocal similarity constraint to improve transmittance estimation accuracy. Firstly, the mist concentration level is calculated from natural scene statistical characteristics of the image. Then the corresponding regularization factor is chosen for transmittance estimation. The local structure similarity constraint is introduced into simplified DCP method to decreasing the uneven transmission estimation error. Weighted quadtree searching method is used to estimate the background light. Finally, the image dehazing model is conducted with the transmission map and background light. The comparative experiments are carried out among the original DCP method, the classical image enhancement method MSRCR and the neural network defogging method PDR-NET. The experimental results show that the mist concentration level prior and nonlocal similarity constraints can optimize the solution of the transmittance. The proposed method has significant effect on increasing the image contrast, reducing mist concentration, keeping texture clarity and edge intensity.
|