Scene-Awareness Based Single Image Dehazing Technique via Automatic Estimation of Sky Area
论文作者：HUI FU , BIN WU, YANHUA SHAO, AND HONGYING ZHANG
School of Information Engineering, Southwest University of Science and Technology, Mianyang 621000, China
论文对于以下三个问题提出了一种新的图像去雾算法：
(1) The enhancement dehazing method is unable to avoid color distortion and/or color shift for it only relies on increasing the color contrast and the brightness of haze images.
(2) Inaccurate estimation of atmospheric light A and transmittance will lead to gradient inversion effects, large image noise, and low efficiency.
(3) For haze images with a large proportion of sky area or water space, it may cause overexposure and mirror reflection and may create a distorted foreground color.
the Gray-level threshold segmentation algorithm 在去雾方面的应用
The modified least-square filter method can help retain the details of the images and suppress noise as well,hence offering a good visual sensory experience.
基于灰度图像阈值分割原理，切割天空区域
Based on the prior information, we can learn that most of the atmospheric light has pixel values between 218 and 223.
使用以上公式对雾图的灰度图像进行天空区域和水域的分割，设置不同的分割阈值，得到的天空区域也是不同，实验表明阈值设置成215可以节省处理时间。
使用skyline算法在天空区域S内寻找大气光值：
改进的最小二乘滤波方法优化传输图像
针对雾度图像的边缘区域经常存在严重的光晕效应，而所选算法应在保留尽可能多的边缘细节的同时去除噪声。提出了使用mean-square算法对传输图像进行优化。
去雾算法的整体思路：
结果对比
结果中仍然存在很多的伪像artifacts.