基于半选择性曝光融合的雾天图像复原算法

Semi-selective exposure fusion method for single image dehazing

基于暗原色理论的相关算法在处理雾天图像时会出现一定色调畸变和曝光问题,严重影响了图像复原的视觉感,提出了一种基于半选择性曝光融合的单幅雾天图像增强方法。该方法首先根据统计试验方法,构建一种基于统计先验的天空区域分割方法,在天空区域内获取大气背景光有效范围。其次,提出一种自适应边界的相对快速双边滤波方法,对透射率图像进行优化,针对性的解决图像的色彩畸变问题。最后,根据大气背景光有效范围,提出半选择性多权重曝光方法,根据不同曝光度的设计,运用多权重方法提取特征,通过融合提高复原的图像视觉效果。为了验证算法在去雾能力及视觉效果的有效性,该文结合多种主观和客观评价方法就行试验,结果证明所提出的算法明显优于已有的算法。

To deal with the color distortion and exposure problems in the dark-channel-prior-based methods,a semi-selective exposure fusion method for single-image dehazing is proposed.According to the statistical prior of hazy images,a sky region segmentation method is first proposed to obtain the effective range of atmospheric light.Then,a fast bilateral filter with an adaptive boundary constraint is proposed to solve the color distortion problem for properly smoothing the transmission.Finally,according to the effective range of atmospheric light and extracting multi-weight features,a selective exposure fusion method is constructed to blend these initial different exposure dehazing images for improving the visual effect of dehazing images.To verify the effectiveness of the algorithm in the ability of dehazing and visual effect,a variety of subjective and objective evaluation methods are carried out,showing that the proposed algorithm is significantly better than the existing algorithms.