摘要
提出一种基于生成式对抗网络的裂缝图像修复方法。在修复过程中,对障碍物所在位置进行信息擦除获得待修复图像。使用生成式对抗网络生成相应的裂缝图像,为待修复图像和生成图像分别覆盖距离加权掩膜,并计算获得修复块。对修复块与待修复图像的拼接图像进行优化获得最终修复结果。实验结果表明,该方法可对裂缝图像进行了准确修复。与传统的修复方法相比,使用该方法修复后的裂缝图像较之前方法峰值信噪比提升了0.6~0.9dB,实现了在有限的裂缝数据集条件下,生成大量还原度较高的裂缝图像。
We proposed a crack image restoration method based on generative adversarial network.In the restoration process,the obstacle location information was erased to obtain the defective image,and the corresponding crack image was generated by using the generated adversarial network.Distance weighting mask was covered for the defective image and the generated image respectively,and the repair block was achieved.Then,the joint image of the repair block and the defective image was optimized to obtain the final restoration result.The experimental results show that the proposed method can repair the crack image accurately.Compared with the traditional restoration method,the peak signal-to-noise ratio of restored image is increased by 0.6 dB to 0.9 dB,which generaes a large number of crack images with high reduction degree under the condition of limited crack data set.
作者
胡敏
李良福
Hu Min;Li Liangfu(College of Computer and Science, Shaanxi Normal University, Xi'an 710119, Shaanxi, China)
出处
《计算机应用与软件》
北大核心
2019年第6期202-208,共7页
Computer Applications and Software
基金
国家自然科学基金项目(61573232,61401263)
关键词
路面裂缝
深度学习
生成式对抗网络
图像修复
Pavement crack
Deep learning
Generative adversarial network
Image restoration
