为了实现基于物理模型的图像复原去雾算法, 文中提出了一种改进的基于暗通道先验的图像去雾算法。介绍了雾天图像退化模型和基于该雾天图像退化模型的几种去雾算法。详细介绍了何恺明提出的基于暗通道先验的去雾算法, 该算法在估计光线传播图时使用的基于导向滤波的软抠图非常耗时, 经过改进, 直接使用景深估计光线传播图, 算法运行时间大大减少。最后, 使用MATLAB对改进的去雾算法进行仿真, 并与原算法的运行时间进行比较。结果显示新方法对光线传播图的估计可靠, 运行时间对比改进前大约下降60%, 实时性大大提高。带有天空的有雾图像去雾后色斑和光晕大幅减少, 取得了很好的效果。改进的去雾算法运行速度快、去雾效果好, 新提出的光线传播图估计方法可靠, 并且去雾过程中得到的光线传播图可以用于其他应用。

To develop an algorithm for haze removal based on the physics model, this paper proposes an improved and fast method for single image haze removal using dark channel prior. First, we introduce the degraded model for describing the formation of a haze image and several algorithms based on this model. Second, we introduce the method of He’s single image haze removal using dark channel prior. The image quality of He’s method is satisfactory, but it is a time consuming method because of refining the transmission map with guide filter. We propose an optimized method based on estimating transmission by scene depth directly and the runtime of the new algorithm decreases a lot. Finally, we realize the algorithm in MATLAB and compare the runtime with the original algorithm. Results demonstrates that the new method provides a reliable transmission estimation and a better image quality with around 40% computation time of He’s method, and the results of haze images with sky are less halos. The optimized method execute fast and the results demonstrate the new method abilities to remove the haze layer as well as provide a high quality transmission estimation as a byproduct of haze removal which can be used for other applications.