GAN-based ancient painting restoration method optimized by GNN with self-attention mechanism

This paper proposes an image restoration method based on Generative Adversarial Networks (GANs), integrating Graph Neural Networks (GNNs) and optimization through the self-attention mechanism. Initially, GNNs are employed for feature extraction from ancient artwork, enabling the full exploitation of its topological and node-based information to generate rich semantic feature representations. These features are subsequently fed into a GAN network enhanced by the self-attention mechanism, which refines the network's focus on key regions and intricate details of the image. This approach allows for more precise reconstruction of missing image parts and yields more natural and realistic restoration outcomes. Experimental results demonstrate that, in comparison to traditional methods, the proposed technique significantly outperforms in the task of ancient artwork restoration, effectively restoring both detailed textures and overall structure. This method holds great potential for practical applications and future advancements in the field.