邻域自适应暗原色先验的单幅图像快速去雾算法

Fast Single Image Haze Removal Algorithm Based on Neighborhood Adaptive Dark Channel Prior

经典暗原色先验去雾理论对于单幅雾天图像有较好的处理效果,但该算法处理halo效应的soft matting技术运行速度较慢.在暗原色先验算法基础之上,针对该算法的不足,提出邻域自适应的暗原色先验去雾的改进算法.算法首先利用暗原色先验原理估算出雾天图像透射率,在计算透射率的过程中引入邻域自适应的思想,在图像边缘的约束下对透射率图进行邻域自适应优化计算,得到了更合理的透射率估计值,使halo效应显著的减少,然后使用导向图滤波对优化后的透射率图再次优化,其中取更小的参数值就可以得到清晰无雾图像,速度得到了加快.实验结果表明,该算法应用到图像去雾处理中,从根本上避免"暗色扩张"现象的发生,而且在去雾效率方面有大大的提高.

The classic theory of dark channel prior achieves good processing results of removing the haze on a single image, but the speed of its soft matting technique for processing the ＂halo＂ effect is relatively slow. Aiming at the deficiency of the algorithm, this paper proposes an adaptive neighborhood optimization algorithm, which is based on the dark channel prior, for removing the haze. Firstly, the algorithm utilizes the dark channel prior to estimate the transmittance of a hazy image. Under the constraint of the image＇s edges, the transmittance of the hazy image is optimized using the neighborhood adaptive mechanism to obtain a more proper estimated transmittance, which can reduce the ＂halo＂ significantly. Then, use the method of guided image filter with a smaller pa- rameter to optimize the optimized transmittance image again. As a result, a clear image without the haze is obtained according to the optimized transmittance image with a faster speed. The experimental results show that when the optimization algorithm is applied to the image haze removal, it can avoid the phenomenon of ＂dark expansion＂ fundamentally. In addition, it can increase the efficiency of removing the haze dramatically.