结合卷积神经网络与动态环境光的图像去雾算法

Image Dehazing Algorithm Based on Convolutional Neural Network and Dynamic Ambient Light

为了有效估计雾霾图像的透射率并改善去雾图像偏暗的问题,提出一种结合卷积神经网络与动态环境光的图像去雾算法。设计了基于卷积神经网络的透射率估计网络,构建包含配对的真实雾霾图像与透射率图像库,对其进行随机块采样,得到配对的雾霾图像块与透射率图像块,并将其作为训练集用于训练透射率估计网络;使用训练好的网络估计雾霾图像的透射率,并进行平滑滤波。考虑到图像成像时光照不均的问题,使用动态环境光替代全局大气光。使用平滑滤波后的透射率和动态环境光进行图像去雾。实验结果表明,该算法不仅可以有效实现图像去雾,而且提高了去雾图像的亮度和饱和度。

To effectively estimate the transmittance of the hazy images and improve the darkness of the fog removal image,an image dehazing algorithm is proposed based on convolutional neural network and dynamic ambient light.Firstly,a transmittance estimation network is designed based on convolutional neural network.Then,an image library containing paired real hazy images and transmittance images is constructed.And randomly block sampling is performed to obtain the paired hazy patches and transmittance patches which are used as training sets for training the transmittance estimation network.After that,the trained network is used to estimate the transmittance of hazy images and then smooth the acquired transmittance.At the same time,considering the problem of uneven illumination of images,dynamic ambient light is used to replace global atmospheric light.Finally,the smooth filtered transmittance and dynamic ambient light are used to restore the images.Experimental results show that the algorithm can not only effectively restore the images,but also significantly improve the brightness and saturation of the restored images.