基于边窗盒子滤波和透射率修正的图像去雾

针对雾线先验去雾算法存在的颜色过饱和现象、图像初始透射率估算不准确等问题，提出了一种基于边窗盒子滤波和透射率修正的图像去雾算法。为了解决初始透射率估算不准确带来的边缘细节信息丢失的问题，首先利用非局部总广义变分（TGV）正则化的方法估算初始透射率，并将二阶的非局部总广义变分（TGV）正则器来作为正则项，以确保对由图像颜色和深度之间的噪声和歧义引起的异常值具有鲁棒性。随后利用边窗滤波算法对初始透射率进行优化，从而实现对图像中纹理信息和边缘信息的保留。最后利用大气散射模型和多角度优化后的透射率复原出无雾的原始图像。实现结果表明，本文算法能够解决图像颜色过饱和与边缘处的细节纹理信息丢失的问题，且无色调偏移和光晕效应。在定性评估上，复原后的图像视觉效果好；在定量评估上，本文算法的去雾后图像的评价指标皆高于基于雾线先验算法。

Image dehazing based on side window box filtering and transmittance correction

Aiming at the problems of color oversaturation phenomenon and inaccurate image initial transmittance estimation in the haze line prior dehazing algorithm, an image dehazing algorithm based on side window box filtering and transmittance correction was proposed. In order to solve the problem of missing edge detailed information caused by inaccurate initial transmittance estimation, firstly the method of the non-local total generalized variation (TGV) regularization was used to estimate the initial transmittance, and the second-order non-local TGV regularized device was used as the regular term to ensure the outliers caused by the noise and ambiguity between image color and depth had robustness. Then, the initial transmittance was optimized by using the edge window filtering algorithm, in this way the texture information and edge information in the image were preserved. Finally, the original image without haze was restored by using the atmospheric scattering model and the multi-angle optimized transmittance. The experimental results show that the proposed algorithm can solve the color oversaturation of the image and the edge detailed texture information loss, and there is no hue shift and halo effect. In the qualitative evaluation, the restored image has a good visual effect, and in the quantitative evaluation, the evaluation indexes of the image after dehazing based on the proposed algorithm are higher than that based on the haze line prior algorithm.