AI-led model aids cultural relic restoration and boosts tourism

A research team led by professor Jiao Licheng from the School of Artificial Intelligence at Xidian University in Xi'an, Shaanxi province, has successfully demonstrated a cultural and tourism resource restoration and generation model in their lab.

After uploading a partial image with a missing corner of the massive mural Chaoyuan Tu, The Assembly of the Gods, from Yongle Palace in Yuncheng, Shanxi province, the system accurately completed the damaged area in just 5 seconds, achieving a restoration effect that is almost indistinguishable to the naked eye.

The innovative achievement, which combines cultural relic restoration with smart tourism services, is injecting strong technological momentum into the high-quality development of Shaanxi's cultural tourism industry.

"To achieve successful restoration, an accurate diagnosis of the mural is essential," said Jiao. "The model provides a diagnostic service, featuring an end-to-end mural damage detection and analysis solution."

"It allows for automated, intelligent processing from damage detection and analysis to restoration plan generation, effectively overcoming the traditional challenges of manual, inefficient, and difficult diagnosis," he added.

At the lab, team member Gao Zihan, a second-year doctoral student, demonstrated the model's diagnostic process by uploading two electronic versions of artifacts into the system.

The model then generated a comprehensive report for the artifacts using blue, purple, white, and black colors to mark types of damage such as peeling, fading, mildew, and cracks, clearly defining the type and severity of the damage. It also proposed targeted solutions such as mildew removal and line supplementation, and suggested specific mineral pigments for filling.

Thereafter, based on the comprehensive diagnostic report, the system can complete missing parts.

"Murals and artifacts from different times have distinct historical characteristics," said Wang Zitao, another second-year doctoral student and team member. "For parts of the mural that cannot be verified, we collaborated with the Huaqing Palace cultural protection center to propose a mural completion method guided by generative mode."

According to the team, the model acts like an excellent student, learning extensively and accurately understanding historical documents and archaeological data. It can organize the learned content and apply it to the mural restoration process, ensuring the completed content aligns with historical facts and aesthetic standards.

During the model's development, the team expanded it with smart tourism services that cater to diverse visitor needs, drawing on their deep understanding of Shaanxi's history and culture.

For example, if a tourist inputs a complex request such as planning a three-day trip to Xi'an with specific dietary and activity needs, the system can quickly generate a custom itinerary that includes an overview, food suggestions, highlights, transportation planning, and precautions.

"Our model is more accurate than other similar products because of our deep collaboration with local cultural and tourism departments, which provides us with more comprehensive and accurate information," said team member Zhang Jie, also a second-year doctoral student. "The model supports both text queries and image uploads and offers multilingual response services."

For instance, uploading an image of the "Elixir" item from the game Black Myth: Wukong allows the model to recognize its similarity to the rhyton cup. It can direct the user to the gold-inlaid agate cup with an animal head displayed in the Shaanxi History Museum and provide detailed historical context.

"The model can offer high-quality, personalized, and accompanying intelligent services for tourists," said the team's academic adviser Liu Xu. "Through deep collaboration with the Shaanxi Tourism Group, we continuously update the model to meet tourist needs, offering pre-trip and on-trip guides, cultural knowledge explanations, and collaborations with local cultural associations for AI-based ancient costume transformations and AI blind box designs."

Relying on more than 30 years of technical expertise in intelligent perception and image interpretation, and with support from cultural heritage protection and tourism planning experts, the model has successfully transitioned from concept to application after more than two years of research, analysis, development, and optimization.

"In the future, the team aims to leverage Shaanxi's rich historical and cultural resources to deepen the integration of technology and culture, expand government-enterprise cooperation, and contribute to building Shaanxi into a world-class smart tourism demonstration area, promoting the innovative dissemination and sustainable development of outstanding historical culture," said Jiao.