改进的Otsu递归分割单幅图像去雾算法研究

Single image dehazing based on an improved recursive segmentation of Otsu algorithm

针对暗原色先验去雾算法对天空区域的透射率估计值偏小问题,提出一种改进的Otsu递归分割图像去雾方法。该方法基于图像分割的思想,利用改进的Otsu递归分割算法准确分离出天空区域,结合暗原色图像灰度归一化值对天空区域透射率估计值进行修正,优化后的透射率更加接近实际值,最后通过大气散射模型得到复原图像。实验结果表明:去雾后的图像保持真实色彩,区域交叉边界处复原自然。文中算法有效解决了天空区域出现色偏和光晕问题,拓宽了暗原色先验算法的适用范围,对户外雾天图像的去雾具有良好的应用前景。

In order to solve the problem of the low estimated transmission value in the sky region by the theory of dark channel prior, this paper proposed a haze removel method based on an improved recursive segmentation derived from Otsu algorithm. Under the thought of image segmentation, the method is accurate to separate sky region via the improved recursive segmentation of Otsu rule, which is combined with the normalized gray values of the dark channel map to amend the estimation for the transmission in the sky region. It optimizes the transmission to get closer to the actual values, then the haze-free image can be recovered by the atmosphere scattering model. Experimental results show that the recovered image preserves the true color and exhibits natural restoration between the boundary region. By effectively dealing with the color distortion and halos appearing in the sky region, our method broadens the application of the dark channel prior in a certain extent,which has a great prospect for processing outdoor foggy image.