基于天空区域分割和边界限制L0梯度最小化滤波的图像去雾算法

Image dehazing based on sky region segmentation and boundary constraint L0 gradient minimization filtering

针对经典的暗原色理论算法在处理带有天空区域的雾天图像时会出现光晕和亮度损失的问题,该文提出基于天空区域分割和边界限制L0梯度最小化滤波的图像去雾算法。首先根据雾天图像的直方图特性分割出天空区域,求解全局大气背景光;其次,根据辐射立方体法则推导出边界限制条件,规整得到初始透射率图像,并运用L0梯度最小化方法对透射率图像进行平滑处理;最后,通过优化的暗原色理论模型求取增强后的图像。通过对算法的有效性、天空区域的失真和细节特征进行分析,发现该算法比改进的暗原色算法处理效果更好。

To deal with the problems of image hue and brightness distortion in the classic dark channel theory algorithm,the image dehazing method based on the sky region segmentation and boundary constraint L0 gradient minimization filtering is proposed here.According to the histogram property of hazy image,the sky regions are segmented to solve the global atmospheric light.Secondly,the boundary constraint equation is firstly used to normalize input image by the radiance cube to obtain the initial transmission which is smoothed by L0 gradient minimization filtering.Finally,according to the optimization of the dark channel theory,a dehazing enhancement image is obtained.Analysis of the effectiveness of the algorithm,the distortion of the sky regions and the detailed feature shows that,the method in this paper is better than the improved dark channel algorithm in term of dehazing visual effect.