含有大片天空区域图像的去雾算法

Image Defogging Algorithm for Images with Large Sky Region

为提高包含天空区域图像的去雾效果,解决暗原色先验原理去雾容易导致天空区域色彩失真以及图像整体亮度较暗的问题,针对含有大片天空的图像,提出一种基于天空分割和色调映射的图像去雾算法.在HSI颜色空间中利用图像众数和图像连通区域提出天空识别算法,分割出天空与非天空区域;然后根据暗原色先验原理分别求取二者透射率,并在天空区域完成大气光值的估计;最后在RGB空间中利用大气散射模型复原图像,并经过改进的自适应色调映射得到最终的去雾图像.采用合成雾图、实景雾图和网络收集雾图进行实验的结果表明,该算法在主观视觉和客观指标方面均能得到质量更好的去雾图像.

Defogging algorithms based on the dark channel prior(DCP) may cause color distortion and darker brightness when they are applied on images with large sky region. To improve the defogging efficiency of DCP based algorithms, a new fog removal method based on sky segmentation and tone mapping is proposed in this paper. This method first designs a sky recognition algorithm in the HSI color space using image modal and image unicom area to separate the sky region from the non-sky region, next calculates the transmission map of the two regions respectively according to DCP and estimates the atmospheric light value using the sky area, then recovers the image in the RGB color space using the atmospheric scattering model. The defogged image is finally produced by further utilizing an improved adaptive tone mapping. The experimental results with synthetic fog map, real fog map and fog map from network show that the images resulted from our method have better objective indicator and fog removal effect compared with some state-of-the-art methods including three deep learning based ones.