Two-Stage Mural Image Restoration with Contextual Aggregation Residuals
Mural image restoration refers to the process of repairing damaged or missing areas in mural images to restore the visual appearance of the image.Addressing issues such as low efficacy and insufficient feature information extraction in existing image restoration methods for mural ima-ges,this paper proposes a two-stage context-aggregated residual adversarial network model for mural image restoration.The entire model consists of a serial network composed of a structure reconstruction network and a color correction network,leveraging context-aggregated residual blocks to extract global and local feature information for image restoration.In the structure recon-struction network,line drawings and self-attention modules are employed to maintain structural sta-bility and global coherence.In the color correction network,SE channel attention modules are uti-lized to enhance the weight influence of inter-channel information transmission,thereby reducing color deviations.Experimental results on a mural image dataset demonstrate that the proposed method outperforms existing restoration methods in both qualitative and quantitative analyses.
deep learningmural restorationcontextual aggregation residualsglobal-local fea-ture
NETL
NSTL
万方数据
