基于神经网络优化的单幅图像去雾算法

Single image dehazing algorithm based on neural network optimization

智能交通建设是我国"十三五"规划的重点课题,图像去雾算法是智能交通系统的重要研究内容之一。提出一种结合暗通道先验和神经网络的单幅图像去雾算法:即先通过改进局部最小值滤波来减小块效应,从而优化粗略传播图;然后结合边缘信息估计大气光值;再通过神经网络代替软抠图算法和引导滤波得到精细化传播图;最后通过大气散射模型恢复出无雾图像。运用目前流行的几种算法以及所提算法分别对有雾图像进行去雾处理,通过去雾效果及分析数据对比来说明所提算法的有效性和适用性。

Intelligent transportation construction is the major subject of China′s 13 th five-year plan.Image dehazing algorithm is one of the important research contents of intelligent transportation system.This paper proposed a single image dehazing algorithm optimized by neural network.The algorithm first optimizes the initial transmission map by improving local minimum filtering to reduce the blocking artifacts;Then,estimating atmospheric light value by combining edge information;The next step,neural network is used to replace the soft mapping algorithm and the guided filtering to obtain the refined transmission map;Finally,the image is restored by the atmospheric scattering model to obtain the haze-free image.Several popular algorithms and the proposed algorithm are applied to restoring images.By comparing experimental results and analysis data,the validity and applicability of the proposed algorithm are verified.