Article: Fusion of BDCN and multi scale U-Net for pattern and line manuscript generation technology in colourful cultural and creative products Journal: International Journal of Information and Communication Technology (IJICT) 2026 Vol.27 No.38 pp.1 - 22 Abstract: This study proposes a CNN-based hierarchical feature decoupling method to address edge breakage and detail loss in digitised line art generation for painted cultural products. The framework incorporates a multi-scale U-Net with a gradient-aware loss function. Experiments show that the proposed algorithm achieves a root mean square error as low as 0.196 and a structural similarity index of 0.982 on the MDBD and BRIND datasets. On the MURAL dataset, the proposed painted cultural and creative product pattern line drawing generation model attains the highest global optimal threshold of 0.886. Compared to pixel difference networks, the search group algorithm, self-attention gating, and random gradient boosting models, the study model demonstrates performance improvements of 12.74%, 13.02%, 9.69%, and 8.23%, respectively. Thus, this study can provide a solution with both technological advancement and cultural adaptability for the digital preservation and design of painted cultural and creative products. Inderscience Publishers - linking academia, business and industry through research

Title: Fusion of BDCN and multi scale U-Net for pattern and line manuscript generation technology in colourful cultural and creative products

Authors: Yuchun Huang

Addresses: School of Design Art, Xiamen University of Technology, Xiamen, 361024, China

Abstract: This study proposes a CNN-based hierarchical feature decoupling method to address edge breakage and detail loss in digitised line art generation for painted cultural products. The framework incorporates a multi-scale U-Net with a gradient-aware loss function. Experiments show that the proposed algorithm achieves a root mean square error as low as 0.196 and a structural similarity index of 0.982 on the MDBD and BRIND datasets. On the MURAL dataset, the proposed painted cultural and creative product pattern line drawing generation model attains the highest global optimal threshold of 0.886. Compared to pixel difference networks, the search group algorithm, self-attention gating, and random gradient boosting models, the study model demonstrates performance improvements of 12.74%, 13.02%, 9.69%, and 8.23%, respectively. Thus, this study can provide a solution with both technological advancement and cultural adaptability for the digital preservation and design of painted cultural and creative products.

Keywords: multi-scale U-Net; CNN; edge detection; painted cultural and creative products; PCCP; line drawing.

DOI: 10.1504/IJICT.2026.153306

International Journal of Information and Communication Technology, 2026 Vol.27 No.38, pp.1 - 22

Received: 07 Aug 2025
Accepted: 18 Nov 2025
Published online: 01 May 2026 *

Title: Fusion of BDCN and multi scale U-Net for pattern and line manuscript generation technology in colourful cultural and creative products

Keep up-to-date