基于暗通道先验的自适应超像素去雾算法

传统基于暗通道先验的图像去雾算法不能有效去除有雾图像在景深突变处的雾点,边界处容易引起光晕效应,对此提出一种基于暗通道先验的自适应超像素去雾算法.首先,在暗通道的获取过程中引入自适应方法判断当前像素邻域内是否具有多个景深物体,若仅存在相同景深物体,则直接求取此像素的暗通道,若存在多个景深物体,则引入超像素分割算法区分不同景深物体,减小景深变化对暗通道获取的影响,以求取更准确的暗通道;然后,估计粗略的透射率,并根据上下文约束细化透射率;最后,通过图像降质的逆过程求解去雾图像.实验结果表明,所提出的算法与暗通道先验单幅图像去雾(DCP)算法、基于边界邻域最大值滤波的快速图像去雾(EMDCP)算法、基于自适应暗原色的单幅图像去雾(ADCP)算法、带边界约束和上下文正则化的高效图像去雾(BCCR)算法相比,可将客观质量综合评价准则提高10%,能够抑制光晕效应,提高有雾图像的视觉效果.

An adaptive superpixel defogging algorithm based on dark channel prior

The traditional defogging algorithm based on dark channel prior can not effectively remove the fog points in the sudden change of field depth, and it is easy to cause halo effect at the boundary. In this paper, an adaptive super-pixel defogging algorithm based on dark channel prior is proposed. Firstly, in the acquisition process of the dark channel, an adaptive method is introduced to determine whether there are multiple depth-of-field objects in the neighborhood of the current pixel. If there are only the same depth-of-field objects, the dark channel of the pixel can be obtained directly. If there are multiple depth-of-field objects, a super-pixel segmentation algorithm is introduced to distinguish different depth-of-field objects, so as to reduce the influence of depth-of-field change on dark channel acquisition and to obtain a more accurate dark channel. Then the rough transmittance is estimated, and the transmittance is refined according to the context constraints. Finally, the defogging image is solved by the inverse process of image degradation. The experimental results show that compared with dark channel prior (DCP), a fast image defogging algorithm based on edge-maximum filter (EMDCP), single image dehazing based on adaptive dark channel prior (ADCP) and efficient image dehazing with boundary constraint and contextual regularization (BCCR) algorithms, the comprehensive assessment criteria index of the proposed method is improved by 10%, which can suppress halo effect and improve the visual effect of foggy images.