AI-based Generative Reconstruction of Intangible Ceramic Patterns for Digital Heritage Preservation

Abstract

Traditional ceramic patterns are an essential component of intangible cultural heritage, embodying aesthetics, region, and cultural identity. However, due to age, insufficient documentation, and the decline of traditional craftsmanship, many of these visual resources are at risk of decline and loss. To address this challenge, this study proposes an AI-based generative framework for the digital preservation and reconstruction of ceramic heritage patterns. This study constructs a dataset of over 2,000 high-resolution ceramic pattern images from museum archives and academic resources, covering representative categories such as blue-and-white porcelain, celadon, and underglaze red. The proposed method utilizes a modified CycleGAN architecture enhanced with an attention mechanism and cultural consistency constraints to capture structural patterns and stylistic authenticity. Experimental results demonstrate superior performance compared to baseline models, improving the Fréchet Inception Distance (FID), Structural Similarity Index (SSIM), and Cultural Consistency Score (CCS). User evaluations involving ceramic experts and design professionals further validate the model's ability to generate visually faithful and culturally authentic patterns. This research highlights the potential of generative AI in preserving intangible cultural heritage while promoting creative innovation in digital ceramic design.