图像去雾中的大气光强度自适应恢复算法研究

Study on Adaptive Recovering Algorithm for Atomospheric Light Vector in Image Dehazing

传统的单幅图像去雾方法中大气光强度仅设定为与图像最亮象素有关的经验值,容易造成去雾后的图像亮度偏暗,且某些区域色彩还原失真等问题。本文提出一种大气光强度自适应恢复算法。首先对图进行分块,根据每个图块的像素在RGB颜色空间分布在同一条线上,可求得大气光强度的方向,然后对大气光强度模值引入一个惩罚因子实现图像去雾,对去雾后图像提出大气光强度模值估计的目标函数,根据图像明暗系数的最大值与透射率的等级无关的约束条件,利用L-BFGS优化从而得到正确的大气光强度模值。实验结果证明该方法可以有效避免大气光强度估值偏差而引起的图像色彩失真,鲁棒性强,去雾后的图像具有更好的色彩还原度和清晰度,更能符合人眼视觉效果。

This paper presented a new method for recovering atmospheric light vector in hazy scenes. Firstly, the haze image was divided into patches, and such patches pixels come into being a line in RGB color space, and the orientation of the atmospheric light vector can be recovered. Then the image was dehazed by using wrong magnitude of atmospheric light vector, and the optimization function was proposed to estimate magnitude of atmospheric light vector. And the restricted condition was added to the optimization function, which means that the max-brightness is independent with different transmission level. So, the magnitude of the atmospheric light vector is recovered by the L-BFGS. Experimental results show that the proposed method can avoid the image color distortion by improving estimation accuracy of magnitude of the atmospheric light vector, and it has robustness. The dehazing images have better color restore degrees and sharpness, which could be more consistent with human visual effects.