ACPAS: an expert-assistance system for authenticating ancient Chinese paintings via LLM-based agents

Abstract

Authentication of ancient Chinese paintings is crucial to protecting and preserving cultural heritage. However, traditional authentication methods rely heavily on expert knowledge and experience, and are difficult to handle unstructured and multimodal information. In addition, the lack of interactive tools also hinders humanities scholars from effectively applying advanced technologies. In this study, we present ACPAS, an intelligent authentication system that enables an expert-led, AI-assisted authentication model. The system uses Large Language models (LLMs) to interpret the needs of experts and assigns them to the corresponding tool modules for processing. It integrates image processing, text retrieval, and structured databases, and employs interactive visualizations to support reasoning. Case studies and user evaluations demonstrate that ACPAS improves efficiency and results in interpretability. It provides a new paradigm for cultural heritage protection and digital humanities research and promotes the deep integration of artificial intelligence and humanities.