基于大气光鲁棒估计的无人机图像去雾方法

A haze removal method for unmanned aerial vehicle images based on robust estimation of atmospheric light

针对无人机(UAV)获取的图像易受雾、霾等天气影响导致图像质量降低的问题,本文提出一种基于大气光鲁棒估计的无人机图像去雾方法。首先,选取具有不同表面反照率的像素块,得到各个图像块的像素直线,利用各条像素直线与大气光共面的性质,估计得到大气光的方向;然后,利用无人机对地成像时图像各像素点的景深相似的特点,定义了图像的全局透射率,通过全局透射率和各像素直线在大气光方向上的投影计算得到大气光幅度;最后,通过对雾天图像模型进行变换得到无雾图像。为使本文方法适用于不同类型的图像,采用了自动调整图像块尺寸和条件阈值等措施来提高方法的鲁棒性。通过真实无人机图像的去雾实验证明,相比现有的图像去雾方法,本文方法在去雾的视觉效果和客观评价指标上都有较大的提升。

Aimed at the problem that the quality of the images acquired by unmanned aerial vehicle （UAV） is easily reduced due to the fog or haze weather, a haze removal algorithm for UAV images based on robust estimation of atmospheric light was proposed. The proposed algorithm selects image patches with differ- ent surface reflectance rate to obtain the pixel line of each patch. Using the properties that all the pixel lines are coplanar with the atmospheric light, the orientation of the atmospheric light vector was calculated. Based on the fact that scene depths of each pixel in the image are similar, the global transmittance is defined. The amplitude of the atmospheric light and the dehazed image are obtained using the global transmittance and projection of the pixel lines on the direction of the atmospheric light. In order to apply this method to different types of images, the measures of automatic adjustment of image block size and condition threshold were adopt- ed to improve the robustness of the algorithm. The experimental results with the real UAV images show that the proposed algorithm has a great improvement in the visual effect and objective evaluation index compared with the existing methods.