基于生成式对抗网络的文物图像超分辨率重建及色彩修复  

Super-resolution reconstruction and color restoration of culturalrelics images based on generative adversarial network

作　　者：朱欣娟[1] 雷倩 吴晓军[2] ZHU Xinjuan;LEI Qian;WU Xiaojun(School of Computer Science,Xi’an Polytechnic University,Xi’an 710048,China;School of Computer Science,Shaanxi Normal University,Xi’an 710062,China)

机构地区：[1]西安工程大学计算机科学学院,陕西西安710048 [2]陕西师范大学计算机科学学院,陕西西安710062

出　　处：《西安工程大学学报》2021年第3期86-92,共7页Journal of Xi’an Polytechnic University

基　　金：国家重点研发计划项目(2017YFB1402102);陕西省重点研发计划项目(2019ZDLSF07-01)。

摘　　要：针对历史久远导致的文物表面暗旧、图像褪色等问题,提出了一种生成式对抗网络的文物图像超分辨(cultural relics super-resolution generative adversarial network,CR-SRGAN)模型。该模型针对图像的退化问题,通过在原始的双三次插值下采样的基础上加上噪音和色彩暗旧处理得到高分辨图像对应的低分辨率图像数据集,然后利用得到的高低分辨率图像训练生成对抗网络,2个子网络不断博弈优化自身性能,最终实现暗旧文物图像的色彩修复和超分辨图像生成。实验结果表明:相比于双三次插值,CR-SRGAN在峰值信噪比(peak signal to noise ratio,PSNR)上平均提高了0.86 dB,在结构相似度(structural similarity,SSIM)上平均提高了0.04,另外从主观来看,褪色图像重建纹理时色彩也得到了一定的修复。A super-resolution generation model for cultural relics image(CR-SRGAN)was proposed in order to solve the problems caused by the long history,such as the dark and old surface of cultural relics and the fading of images.Aiming at the problem of image degradation,the model obtained the low resolution image data set corresponding to the high-resolution image by adding noise and color aging processing on the basis of the original bicubic interpolation down sampling,and then used the obtained high-resolution and low-resolution images to train generative adversarial network.The two sub networks continuously played games to optimize their own performance,and finally realized the color restoration of the dark and old cultural relic image and super-resolution image generation.The experimental results show that,compared with bicubic interpolation,CR-SRGAN has an average increase of 0.86 dB in peak signal to noise ratio(PSNR)and an average increase of 0.04 in structural similarity(SSIM).In addition,subjectively,the color of the faded image is also repaired when the texture is reconstructed.

关 键 词：文物图像 超分辨 色彩修复 生成对抗网络 双三次插值 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]