结合分数阶微分和暗原色先验及Retinex的去雾算法

A Haze Removal Algorithm Combining Fractional Differential,Dark Channel Prior and Retinex

为提高雾霾天气图像的清晰度,提出了一种结合分数阶微分、暗原色先验及Retinex的去雾算法.首先将纹理丰富的雾霾天气下的图像进行分数阶微分,然后将分数阶微分后的图像作暗原色先验处理,并根据暗原色景深图信息计算其在图像中的Retinex尺度,最后对分数阶微分后的图像进行Retinex变换得到结果图像.对一系列雾霾天气下图像的测试结果表明:文中算法能够有效提高雾霾天气中模糊图像的清晰度,减少Retinex的光晕现象;与现有的多尺度Retinex及暗原色先验算法相比,对于纹理丰富及场景深度差异较大的雾霾天气下的图像,文中算法既能保持良好的增强效果,又可以加快运行速度.

In order to improve the clarity of haze images, this paper proposes a haze removal algorithm combining the fractional differential, the dark channel prior and the Retinex. In the algorithm, a high-texture haze image is processed first through the fractional differential and then through the dark channel prior. Moreover, on the basis of the depth map obtained through the dark channel prior, the Retinex scales are calculated in each part of the pro- cessed image. Finally, the image enhancement result is got by performing the Retinex transform of the image after the fractional differential operation. The test results of a number of haze images show that the new algorithm can effectively improve the clarity of haze images with less Retinex halo phenomena, and in comparison with the existing dark channel prior and multi-scale Retinex algorithms, it has a higher processing speed and a better image enhance- ment effect for the haze images of the high texture and great scene depth difference.