一种结合物理模型和景深估算的图像去雾算法

A Dehazing Method Combining Physical Model and Depth Map Estimation

图像去雾是图像处理和计算机视觉领域中的一个重要课题。为解决现有去雾算法容易造成颜色失真的问题,提出一种结合物理模型和景深估算的去雾方法。首先,采用基于四叉树细分的方法估算大气光值;然后,通过颜色衰减先验建立线性模型,根据景深获得自适应的大气散射系数和透射率;最后,利用大气散射模型得到无雾图像。实验结果证明:在视觉效果方面,去雾算法效果明显。在客观指标方面,利用He、Meng、Zhu方法及文中方法对5幅不同图像进行去雾,文中均方误差(MSE)分别比其他3种图像去雾算法中最低的值降低了 13%,78%,23%,25%,75%,峰值信噪比(PSNR)则分别提高了 4%、31%、5%、8%、31%。

Image dehazing is an important topicin the field of image processing and computer visualization.An improved image dehazing method based on color attenuation prior is proposed in this paper to address the problem of the existing dehazing algorithms that are prone to color distortion.First,a method based on quad-tree subdivision is used to estimate the atmospheric light.Then,a linear model is established using color attenuation prior,and adaptive atmospheric scattering coefficients and a transmission map are obtained according to the map depth map.Finally,a haze-free image is obtained from the atmospheric scattering model.To compare the effect of the proposed method with other existing methods,five different images are dehazed with the proposed algorithm and three other methods(He,Meng and Zhu methods).In the experiments,the proposed algorithm shows better dehazing effect than the other three methods.The mean square error(MSE) of the five images with the proposed method decreased by 13%,78%,23%,25%,75%,respectively,compared to the lowest MSE of the other three methods.The peak signal-to-noise ratios(PSNR) of the five images with the proposed method increased by 4%,31%,5%,8%,31%,respectively,compared to the highest PSNR of the other three methods.