Haze has an impact on the quality of the image. Single image dehazing is a challenging ill-posed prob-lem. The traditional dehazing methods have some problems, such as color distortion and limited application scope. To overcome these problems, we propose a generative adversarial mapping nets(GAMN) algorithm for image dehazing. In the training, an adversarial learning mechanism between the generative networks and the discrimina-tive networks was used to obtain the optimal solution of parameters. In the testing, the trained generative net-works can translate the haze related features to the medium transmission by multilayer Perception, the medium transmission is related to the depth and help to complete dehazing. Experimental results show that the proposed algorithm is closer to the real color compared with the state-of-the-art method. It can restrain noise and dehaze clearly.%雾霾常会影响获取图像的质量单幅图像去雾是一个具有挑战性的不适定问题.针对传统的去雾方法存在去雾结果颜色失真、适用范围局限等问题提出一种基于深度网络的去雾算法——生成对抗映射网络的多层感知去雾算法.在训练阶段中利用生成对抗映射网络里判别网络与生成网络间对抗式训练机制保证生成网络中参数的最优解;在测试还原过程中先提取有雾图像中雾气相关特征并利用训练得到的生成网络对提取特征进行多层感知映射进而得到反映雾气深度信息的透视率最终运用得到的透视率实现了图像去雾.实验结果表明与同类算法相比该算法能较好地还原出场景中目标的真实色彩并抑制部分噪声去雾效果明显.
展开▼
