A Digital Restoration Method for Damaged Ancient Paintings Based on Hybrid Domain Pix2pix Model

Due to the numerous factors that can cause damage to ancient paintings, digital restoration work is very challenging without prior knowledge. To fully utilize the deep semantic features of images during the digital restoration process, an improved digital restoration model for damaged ancient paintings based on the Pix2Pix model - Hybrid Domain Pix2Pix model is proposed. This model builds on the Pix2Pix framework and introduces a multi-scale attention mechanism to enhance the integration of macro structural information and micro texture details. Meanwhile, a Swin Transformer module is added in order to improve feature extraction capabilities. Additionally, to better capture high-frequency details in images, the original U-Net is replaced with an FFC-UNet network structure. Furthermore, by incorporating perceptual and style losses alongside the mean absolute error (L1 loss) and adversarial loss, the performance of the generator in restoring image details and overall coherence is enhanced. Experimental results show that the Hybrid Domain Pix2Pix model achieves better visual effects and significant improvements across multiple objective evaluation metrics. These results demonstrate the effectiveness of the proposed model and provide a solution for the digital restoration of damaged ancient paintings.