Visual Design from Cultural Heritage through Inpainting-based Restoration

Jungmin Lee, Jongwon Choi; Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) Workshops, 2025, pp. 4432-4437

We explore an inpainting-based restoration framework for generating commercial designs from cultural heritage paintings. Cultural heritage paintings often exhibit accumulated damage, including surface contamination, color alteration, and physical deterioration, which compromises the quality of generative design outputs when processed directly. Our approach addresses this challenge by preprocessing damaged paintings through inpainting techniques before applying generative design algorithms. We employ damage-specific restoration strategies that include cleaning to remove surface contamination, coloring to restore the paint layer, and pattern extraction to isolate decorative elements. Through comparative analysis of designs generated from damaged originals and restored paintings, we demonstrate that inpainting preprocessing significantly enhances visual quality while preserving cultural authenticity. The restored designs show improved commercial viability and meet contemporary design market standards. This approach bridges historical preservation with modern design innovation, offering market-ready designs that maintain rich cultural resonance in diverse applications.