基于超像素和暗通道先验的图像去雾复原方法

Image Dehazing Method Based on Superpixels and Dark Channel Prior

基于暗通道先验的图像去雾算法因其简单有效而得到广泛应用。经典算法中由于透射率图估计存在块效应而需要对透射率图进行滤波细化。本文提出了一种基于超像素分割和暗通道先验的透射率估计方法,用超像素分割获得的不规则区域替代经典算法中的固定窗口,从而避免了透射率图估计的块效应问题,并采用松弛策略以获得鲁棒性和准确性更好的透射率结果,无需进行另外的滤波细化步骤即可完成图像去雾复原。实验结果证实了本文所提方法的有效性。

Image defogging algorithms based on dark channel prior(DCP)are widely used because of its simplicity and efficiency.Traditional DCP-based algorithms may yield to block artifacts of the estimates of the transmission map so that additionalfilter has to be employed for refinement.In this paper we propose a method of estimation of the transmission map based on super-pixel segmentation and dark channel prior.The fixed windows used in traditional algorithms in estimation of transmission mapare replaced by the irregular regions obtained by super-pixel segmentation,which can effectively avoid the block artifacts in estimation of the transmission map.A relaxation algorithm is proposed to obtain better estimation of the transmission map in both robustness and accuracy.The image defogging can be completed without any additional filtering processes.Experimental results show the effectiveness of the proposed method.