单幅夜间弱照度雾霾图像的复原算法

摘要

针对单幅夜间弱照度雾霾图像的复原问题,提出一种算法.首先将原图像分为纹理层和结构层,对结构层的照射光初步估计之后再优化;然后根据Retinex理论将结构层与优化后的照射光的比值作为反射层,对其高亮区域的过增强及暗区域的噪声进行抑制,再去雾处理;接着将已优化的照射光取反作为透射率的估计值、对夜间环境光用求取局部均一的方式进行估计,再根据大气散射模型求出复原的结构层;最后,将复原的结构层与优化后的纹理层叠加为最终的复原图像.与现有主流算法的主客观比较和分析结果表明,该算法的复原结果具有噪声低、纹理细节丰富和色彩恢复度高的优点.%In this paper,a algorithm is proposed to aim to the challenge of the restoration algorithm of the single nighttime weakly illuminated haze image.Firstly the original image is decomposed into two layers:structure and texture layers.After the initial estimation to the illumination of the structure layer,the illumination is refined.The ratio of the structural layer to the refined illumination is used as the reflection layer based on the Retinex theory.After the suppression to the over-enhancement highlight area and dark area with noises,haze removal processing can be applied.The estimated transmission is the inversion of the optimized illumination.After the ambient light is estimated by means of calculating the local homogeneity,the restored structure layer can then be obtained based on the atmospheric scattering model.Finally,the optimized texture layer is added back to the restored structure layer to generate the final restored image.Comparison with the existing advanced algorithms via objective and subjective analysis,the restored results of the proposed algorithm have the advantages of less noises,clearer texture details and higher color restored degree.