Self-supervised refined edge detection for complex Tibetan patterns via bimodal fusion
Abstract
This paper proposes a self-supervised edge detection method, FCL-SSED, to address challenges in Tibetan pattern analysis, such as dense lines and label scarcity. We construct a 5-category dataset and utilize pseudo-labels for iterative training. To enhance fine-grained feature extraction, a Bimodal Weighted Edge Fusion (BWEF) strategy is introduced. Furthermore, an Adaptive Multi-Strategy Fusion (AMSF) method is proposed to generate high-quality pseudo-ground truth for quantitative evaluation, complemented by no-reference metrics. Experimental results demonstrate that the method achieves superior performance, providing a novel technical solution for the digital preservation and inheritance of Tibetan patterns.
Fields of Study & Subjects
Image and Object Detection Techniques
Advanced Neural Network Applications
Image Processing and 3D Reconstruction
Publication details
- Journal / source
- The Imaging Science Journal
- Volume
- 74
- Issue
- 6
- Pages
- 639–659
- Publication date
- 24 February 2026
- Language
- EN