生成对抗映射网络下的图像多层感知去雾算法  

Generative Adversarial Mapping Nets with Multi-layer Perception for Image Dehazing

作　　者：李策[1,2] 赵新宇 肖利梅[1] 杜少毅 Li Ce;Zhao Xinyu;Xiao Limei;Du Shaoyi(College of Electrical and Information Engineering, Lanzhou University of Technology, Lanzhou 730050;The School of Electronic and Information Engineering, Xi’an Jiaotong University, Xi’an 710049)

机构地区：[1]兰州理工大学电气工程与信息工程学院,兰州730050 [2]西安交通大学电子与信息工程学院,西安710049

出　　处：《计算机辅助设计与图形学学报》2017年第10期1835-1843,共9页Journal of Computer-Aided Design & Computer Graphics

基　　金：国家自然科学基金(61365003);甘肃省基础研究创新群体项目(1506RJIA031);中国博士后科学基金(2014M550494)

摘　　要：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.Haze has an impact on the quality of the image.Single image dehazing is a challenging ill-posed problem.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 discriminative networks was used to obtain the optimal solution of parameters.In the testing,the trained generative networks 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.

关 键 词：生成对抗映射网络 多层感知 雾气相关特征 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]