Degradation-Aware Hybrid Attention Transformer with Fidelity Constraints for Cultural Heritage Image Super-Resolution
Year: 2026
Type: article
Abstract: Digital preservation of cultural heritage images requires resolution enhancement that maintains agreement with existing digital records. However, conventional image super-resolution methods are mainly developed for natural images and may suffer from limited degradation adaptability, color inconsistency, and unsupported detail generation when applied to heritage images. To address these challenges, this paper proposes a Heritage-Faithful Hybrid Attention Transformer (HF-HAT) framework for cultural heritage image super-resolution. HF-HAT introduces a heritage degradation encoder to learn input-dependent degradation representations and adaptively modulate Transformer-based reconstruction. Furthermore, a multi-dimensional heritage fidelity constraint strategy is developed to preserve structural, chromatic, and textural consistency between reconstructed images and digital references. A reconstruction reliability head predicts relative spatial reconstruction-error scores from shared features and identifies regions with larger estimated errors. Experiments on MuralDH show improved reconstruction accuracy and closer structural and color agreement with paired high-resolution digital references. These findings support reference-guided super-resolution for cultural heritage digital images.
Source: Applied Sciences
DOI: 10.3390/app162211279
Authors: Kangda Cheng, Xiaobao Li, Jinlong Liu
Institutions: Jimei University, Harbin Institute of Technology
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $2,896
Funder: Jimei University
Award: G9001338681
OpenAlex ID: W7226014846
