Descripción:
Digital preservation of cultural heritage requires specialized methodologies that balance technical restoration quality with historical authenticity. This work presents HistoRestorer, a Vision Transformer architecture specifically adapted for Chinese historic mural restoration. We introduce three pioneering cultural metrics: Historical Authenticity Index (HAI), Cultural Pattern Similarity (CPS), and Chromatic Fidelity Score (CFS), representing the first computational formalization for historical authenticity evaluation. Our approach employs a lightweight transformer architecture (867K parameters) achieving 181 img/s throughput on the DhMural1714 dataset (1,714 authentic images with 8,000 augmented samples). Comprehensive evaluation using 15+ validation techniques demonstrates stable training convergence with peak HAI of 0.584 and final HAI of 0.572, though a train-validation gap (0.316±0.084) indicates areas for improvement. Ablation study confirms the critical importance of the transformer component (-0.329 HAI without it). This work establishes the first computational framework for cultural authenticity metrics and provides a rigorous evaluation protocol for heritage computing applications.
Tipo de publicación: Conference Paper
Publicado en: 2025 IEEE VIII Congreso Internacional en Inteligencia Ambiental, Ingenieria de Software y Salud Electronica y Movil (AmITIC)
Autores- Mendoza Valdés, José Longino
- Quesada-López, Christian
- Méndez Porras, Abel
Investigadores del CITIC asociados a la publicación
Dr. Christian Quesada-López
Dr. Abel Méndez Porras
Proyecto asociado a la publicación