在雾、霾等天气条件下, 大气粒子的散射作用使环境的能见度偏低, 视觉系统采集到的图像严重降质。基于暗通道先验的图像复原方法因其去雾效果自然、约束条件少, 且易于实现等优点而受到广泛关注。但是, 该方法的去雾效果受尺度(暗通道的求解半径)影响很大, 对于不同场景的图像, 不存在一个普遍适用的最优尺度。针对该问题, 文中提出一种尺度自适应方法, 根据图像的颜色和边缘特征自适应地调节暗通道的尺度范围, 得到像素级的暗通道求解尺度, 兼顾大尺度求解色彩失真小和小尺度求解"光晕"失真小等优点。此外, 针对暗通道去雾方法会使天空光估计点落到前景区域的问题, 提出了一种改进的天空光估计方法, 可使估计点鲁棒地落到与其物理意义相符的背景区域。对多种雾化场景图像的处理结果表明: 文中方法适应性强、去雾效果自然, 且对比度提升显著。

In fog and haze weather conditions, scattering of atmospheric particles greatly reduces the outdoor visibility. Images captured by vision system suffer from serious degradation. Haze removal using the dark channel prior is considered to be a good solution due to its advantage of simple implementation and pleasing result with little constraint. While the selection of scale(radius of patch size) determines quality of the recovered image. For different scenes, there is no generally applicable scale. To solve this problem, in this paper, a scale adaptive method was proposed. It adjusted the range of scale adaptively according to features of color and edge, and get the pixel-level scale of dark channel. Proposed method has both advantage of little color distortion and little "halo" artifacts. In addition, an improved method of atmospheric light estimation was proposed. By this approach, the estimation point robustly fell into the background region, and that was physically sound. Experimental results on a variety of outdoor hazy images demonstrate that the proposed method is general applicable. The method also achieves pleasing results of haze removal with good color atmosphere and higher contrast.