BROWSE
Detail
Conference Paper
딥러닝 특징점 매칭 기반의 고미술품 초고해상도 자동 스티칭 및 형태 보정 파이프라인
Cited - time in 
Share 
Share
- Issue Date
- 2026-06
- Citation
- 대한전자공학회 학술 대회 (하계) 2026, pp.1-5
- Publisher
- 대한전자공학회
- Language
- Korean
- Type
- Conference Paper
- Abstract
- The digitization of cultural heritage and antique art requires gigapixel stitching technology that merges hundreds of macro images to preserve precise textures. However, existing traditional feature point extraction methods frequently cause matching errors in texture-deficient Hanji or repetitive brushstroke areas, leading to severe out-of-memory issues when processing ultra-large images. In this paper, we propose a fully automated stitching pipeline that combines a Transformer-based Global Context Matching Model, Adaptive Canvas Extension, and Multi-Resolution Spline Blending (Enblend). Furthermore, we demonstrate the stability and visual integrity of enterprise-grade large-scale stitching by introducing a sequential mapping algorithm that prevents the computational explosion (O(N²)) of large-scale grid data and an automatic horizontal correction system based on the Hough Transform. Ultimately, this study presents a robust, open-source framework that achieves full automation of gigapixel panorama construction without relying on commercial software.
- KSP Keywords
- Automated stitching, Context matching, Correction system, Cultural Heritage, Extraction method, Feature point extraction, Global context, Hough Transform, Mapping algorithm, Matching model, Open source