基于亮区域暗通道矫正和加权聚合引导滤波

针对暗通道先验在天空等亮区域失效和引导滤波容易导致边缘模糊的不足，提出了一种高效的去雾算法.该算法提出一种新颖的亮区域自适应分割与校正方法，基于自适应的相对景深阈值分割亮区域，采用饱和度和灰度值校正亮区域的暗通道.然后利用加权聚合引导滤波代替引导滤波细化初始透射率，解决引导滤波引起的边缘模糊问题.最后，提出一种有效的亮度校正方法，将复原图像转化到HSV色彩空间，对亮度进行均衡化处理，使用相对雾浓度均值作为权值，对均衡化前后的结果进行线性加权得到最终的复原结果.实验结果表明，与经典算法对比，所提算法亮区域分割准确，复原图像纹理清晰，去雾彻底，复原结果的峰值信噪比、平均梯度与信息熵的最大提升分别为34.46%、99.49%和21.18%.

An image dehazing algorithm based on dark channel correction and weighted aggregate-guided filtering

A new efficient dehazing algorithm is proposed to address the failure of the dark channel prior method in bright areas such as the sky and the blur problem at the edges caused by guided filtering. In this algorithm, a novel adaptive segmentation and correction method is first proposed; the bright areas are segmented based on the relative depth of field threshold, and the value of saturation and grey level are used to calibrate the dark channel in the bright areas. Then, the weighted aggregate guided filtering instead of guided filtering is adopted to refine the initial transmittance to solve the problem of edge blur caused by guided filtering. Finally, an effective brightness correction method is proposed; the restored image is converted to HSV color space to equalize brightness, and the results before and after brightness equalization are linearly weighted to obtain the final result by using relative fog density average as the weigh. The experimental results shows that the proposed algorithm can segment the bright area accurately, restore the image texture clearly and remove the fog thoroughly; the maximum improvement of peak signal to noise ratio, average gradient, information entropy are 34.46%, 99.49%, 21.18%, respectively, by comparing with the previous results.