Image Dehazing Based on Accurate Estimation of Transmission in the Atmospheric Scattering Model
Image Dehazing Based on Accurate Estimation of Transmission in the Atmospheric Scattering Model
复制标题
DOI：
10.1109/jphot.2017.2726107
复制
发表时间：
2017-07
影响因子：
2.4
通讯作者：
Guoling Bi;Jianyue Ren;Tianjiao Fu;Ting Nie;Changzheng Chen;N. Zhang
中科院分区：
文献类型：
--
作者：
Guoling Bi;Jianyue Ren;Tianjiao Fu;Ting Nie;Changzheng Chen;N. Zhang
图像去污是计算机视觉应用中一项极具挑战性和极具吸引力的技术。暗通道优先(DCP)是近年来被认为是一种有效的去雾技术。然而，DCP的失效会导致不可靠的传输估计，导致不准确的颜色信息恢复、光晕伪影和块效应。通过对室外无霾/雾霾图像的观测，提出了一种能够反映场景亮度信息和光线反射能力的亮度图，并给出了DCP与亮度图之间的数学模型。该算法能够有效地补偿DCP，准确估计透射图，自适应地获取全局大气光，实现图像的自动分割。利用多尺度导引滤波细化透射图，能够避免场景深度突变时的晕影伪影。通过一系列的实验证明，该算法可以获得高质量的无雾霾图像，具有丰富的可分辨细节、低的颜色失真和较少的晕影伪影，其性能优于或接近于目前最先进的四种雾霾去除算法。
Image dehazing is a challenging and highly desired technology in computer vision applications. The dark channel prior (DCP) has been considered to be an efficient dehazing technique in recent years. However, the invalidation of DCP can induce unreliable estimation of transmission, resulting in inaccurate color information recovery, halo artifacts, and block effect. In this paper, a novel brightness map is proposed based on the observation on outdoor haze-free/haze images that can reflect the brightness information and the light reflection ability of the scene, furthermore, the relationship between DCP and the brightness map is given in mathematical model. The proposed algorithm can compensate for the DCP effectively, estimate the transmission map accurately, get the global atmospheric light adaptively and segment the image automatically. Using multiscale guided filter refine transmission map, the halo artifacts are able to be avoided in the scene depth of a sudden change. A series of experiments are additionally implemented to demonstrate that the proposed algorithm can obtain high-quality haze-free images with abundant distinguished details, low color distortion, and little halo artifacts that can outperform or be comparable with four state-of-the-art haze removal algorithms.