Abstract: Dark channel prior often gives a smaller estimation value of transmission in bright areas and sky regions, resulting in brightness dim and color distortion of restored image. So this paper proposes a single image dehazing method. Firstly, a smooth dark channel is defined to describe the central tendency of three color channel values when the mean filter is used to calculate the dark channel instead of the minimal filtering. The smoothing transmission is calculated roughly by mean filter, and then is refined through guided filter. The atmospheric light is estimated from the most haze-opaque pixel. The method avoids to the halo effect and black spot. The values of image luminance and atmospheric light are compared. When in a range of greater than or less than atmosphere bright, the pixel value is seen as a point of light areas, and the transmittance is corrected. The method makes the obtained transmission more precise, details of restored image more distinct. The experimental results show that the proposed algorithm can effectively solve the problem of the image distortion in the large and bright areas, and the restored image has high brightness and contrast.

Key words: image dehazing, dark channel prior, mean filter, guided filter

摘要： 针对暗原色先验在明亮区域和天空区域透射率估计值偏小，致使复原图像亮度偏暗、颜色失真等问题，提出了一种新的图像去雾算法。在计算暗通道函数时，定义了一类平滑暗通道对3个颜色通道值的集中趋势进行描述，则该区域像素点的暗通道的值为其三原色通道的平均值，代替原来的最小值。使用均值滤波得到平滑的粗透射率，再通过引导滤波对透射率进行细化处理，进而估计全球大气光值，有效地去除了光晕效应及黑斑效应。将图像像素的亮度值与全球大气光值进行比较，对处在一定范围内大于或小于大气光值的像素点作为明亮区域的点，并对该点的透射率进行修正，使求得的透射率更为准确，复原后的图像细节更加清晰。实验结果表明，该算法能有效解决大面积明亮区域图像失真的问题，复原后的图像也具有较高的亮度和对比度。

关键词: 图像去雾, 暗通道先验, 均值滤波, 引导滤波