针对暗通道先验去雾算法在处理单幅图像去雾时计算复杂度高且对于天空等高亮区域有局限性而易产生失真的问题, 从暗通道模型出发, 提出首先利用双暗通道拟合进行透射图估计, 然后采用K均值聚类算法对有雾图像进行区域分类之后再针对天空区域估计出大气光强度的算法。该算法增强了图像的细节信息, 并大大降低了计算复杂度, 且提升了大气光强度估计值的准确性, 有效抑制了高亮区域的失真。主观和客观评价表明, 该算法能够取得比传统算法更好的去雾效果。

The traditional single image dehazing algorithm based on dark channel prior has high computational complexity, and is prone to distortion with the sky and other highlighted areas. Based on a prior theory of dark channel, in this paper, first multi-dark-channel fitting was used to get the transmission map, then the K-means clustering algorithm was used to obtain a non-sky area. And then, the atmospheric light intensity in the sky area was estimated, which enhances the image detail and greatly reduces the computational complexity, and the accuracy of the atmospheric light value is more accurate, and also the distortion in the highlighted area is effectively suppressed. Through the subjective observation and objective evaluation, this algorithm can obtain good defogging effect compared with the classic dark channel algorithm.