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.

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. [ABSTRACT FROM AUTHOR]