Image Super-Resolution Reconstruction Techniques in Digital Restoration of Artworks

Abstract

Enhancing image quality during the digital restoration of artworks presents a critical technical challenge. Addressing the issues of low integration and limited processing efficiency in existing digital restoration systems, this paper proposes an end-to-end integrated system featuring multi-module collaboration. It organically combines the Deep Attention Network (DBAN) architecture with functional modules for image processing, quality assessment, and workflow management. The deep learning-based DBAN model fuses texture, edge, and color features unique to artworks, achieving high-quality super-resolution reconstruction with a peak signal-to-noise ratio (PSNR) of 33.8 dB and a structural similarity (SSIM) of 0.943. Through a unified system architecture design, it integrates functional chains including data preprocessing, model training, distributed inference, and result evaluation, while enabling seamless integration with existing digital art archive management systems. Employing a microservices architecture, the system integrates distributed processing, model compression, and service deployment components to achieve a processing capacity of 122 images per second. Field tests at multiple art restoration institutions demonstrate that this integrated system reduces restoration time by 45% while maintaining reconstruction quality, providing a comprehensive technical solution for large-scale digital restoration of artworks.