Damaged Artwork Restoration Using Mask R-CNN and CA-GAN

Art restoration plays a vital role in preserving cultural heritage and maintaining the beauty and history of damaged artwork. Traditional methods require experts, making the process expensive and time-consuming. We aim to provide a digital restoration of damaged artworks using deep learning, employing a combination of Mask R-CNN for automated mask generation and a CA-GAN (Contextual Attention Generative Adversarial Networks) for image inpainting. This method simplifies the restoration workflow by first employing a pre-trained Mask R-CNN model to detect and generate masks for damaged regions in the artwork. These masks precisely identify areas requiring restoration, ensuring minimal manual intervention. Subsequently, CA-GAN utilizes its contextual attention mechanism to learn spatial dependencies and texture patterns from undamaged regions, enabling the seamless inpainting of masked areas. The model is evaluated based on PSNR and SSIM metrics achieving PSNR of 35.10 dB and an SSIM of 0.952. The model preserves the artistic style and intricate details of the original painting by synthesizing high- quality textures and ensuring smooth blending with the surrounding regions.