Variational Model for Digital Restoration of Monumental Painting
Abstract
Variational Model for Digital Restoration of Monumental Painting
Incoming article date: 02.11.2025The paper addresses the problem of digital restoration of monumental painting through the reconstruction of lost color fragments. A variational model is proposed, illustrating a two-stage approach: segmentation of the image with identification of damaged regions using a convolutional neural network based on the U-Net architecture, followed by color reconstruction in the selected areas using a convolutional autoencoder in the CIELAB color space. The specific features of applying the discussed neural networks, the data preparation workflow, and the training parameters are described. The results demonstrate that the proposed approach provides reliable detection of defective areas and high accuracy of color restoration while preserving the artistic style of the original. The limitations of the method and potential directions for further development are also discussed.
Keywords: variational model, monumental painting, digital restoration, image segmentation, convolutional neural network, color reconstruction, convolutional autoencoder, CIELAB color space