Denoising of oracle bone rubbing images via a residual network based on base-residual decomposition

Abstract The optimized residual denoising network demonstrates effectiveness in addressing the issue of noise throughout the digitalized documentation process of cultural relics in oracle-bone inscriptions. It integrates base-residual decomposition processing, and residual learning, and batch normalization. On the one hand, the base-residual decomposition enables the proposed residual network to process the two components separately, stabilizing model convergence. On the other hand, it can effectively balance denoising performance and detail retention without introducing obvious artifacts. Experimental results show that the proposed model outperforms the BM3D, IRCNN, and TL1 baselines considered in this study in most visual quality and objective metrics. It is of practical value for the digital protection and intelligent processing of oracle-bone inscription cultural relics.