Tomb Mural Image Enhancement based on Improved CycleGAN
Jing Zhao · 2022
For the problem of fading or discoloration, the E-CycleGAN model is proposed to realize the digital restoration of tomb murals in color. Specifically, first, StyleGAN3 network is used to generate the tomb mural face; then use empty convolution to replace the original convolution in CycleGAN, to expand the detail information of the mural face, and finally, based on CycleGAN Loss, add Identity Loss to ensure that the content of the original image does not change. Repair was performed on the self-built tomb chamber mural face dataset, and the experimental results showed that the NIQE index decreased by 1.72% on average. It proves that the network has obtained better restoration results in the color restoration of the tomb murals.