基于像素级图像融合的单幅图像去雾算法  

Novel Algorithm for Single Image Dehazing Based on Image Fusion at Pixel Level

作　　者：刘言言[1] 沈东升[1] 林梦雷[1] 

机构地区：[1]闽南师范大学计算机学院,福建漳州363000

出　　处：《闽南师范大学学报（自然科学版）》2016年第1期46-53,共8页Journal of Minnan Normal University：Natural Science

基　　金：福建省自然科学基金(2013J01028);福建省计算机应用技术和信号与信息系统研究生教育创新基地项目([2008]114号);福建省数学类研究生教育创新基地(313009);福建省中青年教师教育科研项目(科技)(JA15302)

摘　　要：单幅图像去雾是计算机视觉领域的一个重要研究课题,基于图像融合思想,提出一种新的单幅图像去雾算法.首先计算大气光和中值暗原色先验的差值绝对值来判断有雾图像中是否含有明亮区域,获得对天空、白色建筑物等明亮区域透射率更精确的估计,并通过该透射率计算第一幅待融合图像;然后利用大气散射模型的一般形式,求解出第二幅待融合图像;最后,通过计算融合系数,将两幅去雾图像进行像素级融合,得到最终的去雾图像.该方法可以有效的改善天空区域颜色失真,去除Halo效应.实验结果表明,所提方法能较好的实现去雾,并保留图像细节和结构信息.Single image dehazing is an important topic in the computer vision. A novel algorithm for single image dehazing based on image fusion at pixel level is proposed. Firstly, a transmission map, which is more accurate on bright areas in the image, is obtained by computing the discrepancy between the median dark channel prior and the skylight, and the first input image for fusion is derived by the transmission map. Secondly, the other input image for fusion is derived using general atmospheric scattering model. Finally, the dehazed image is restored based on image fusion at pixel level using the above input images. The proposed method can effectively improve the sky areas distortion, remove the halo effect. Experiments demonstrate that the proposed method would better remove the haze, preserve the details and structure information of the images.

关 键 词：图像去雾 图像融合 中值暗原色先验 大气散射模型 Halo效应 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]