Article: Intelligent generation algorithm for digital image artworks based on decoupling representation and content-aware Journal: International Journal of Information and Communication Technology (IJICT) 2026 Vol.27 No.57 pp.23 - 50 Abstract: Focused on calligraphy, this research addresses style transfer distortion and inadequate compositional aesthetics in AI-generated art. We propose an algorithm that integrates decoupling representation learning with content-aware layout modelling. A dual-encoder architecture separates character structure and brushstroke style features, enabling precise and controllable style transfer via dynamic instance normalisation. A visual-linguistic bimodal network with hierarchical spatial modules is introduced to model relationships at the character, line, and global levels. The proposed method achieves a style similarity of 0.751, a content preservation PSNR of 36.9 dB, and an 8%-16% improvement in cross-font generalisation accuracy on unseen characters. For layout generation, the framework maintains a line-spacing fluctuation coefficient of 0.032, achieves a layout aesthetics score of 4.8, and demonstrates strong long-text stability with a cross-page style consistency of 0.94. Ablation studies further confirm the effectiveness of the dynamic weight adjustment mechanism, achieving an optimisation efficiency of 0.98. This work addresses key technical bottlenecks in digital calligraphy generation, providing a practical tool for cultural heritage preservation and a transferable framework for other structured art generation tasks, thereby advancing the integration of artificial intelligence with traditional arts. Inderscience Publishers - linking academia, business and industry through research

Title: Intelligent generation algorithm for digital image artworks based on decoupling representation and content-aware

Authors: Liyuan Zhang

Addresses: Lu Xun School of Arts, Yan'an University, Yan'an, 716000, China

Abstract: Focused on calligraphy, this research addresses style transfer distortion and inadequate compositional aesthetics in AI-generated art. We propose an algorithm that integrates decoupling representation learning with content-aware layout modelling. A dual-encoder architecture separates character structure and brushstroke style features, enabling precise and controllable style transfer via dynamic instance normalisation. A visual-linguistic bimodal network with hierarchical spatial modules is introduced to model relationships at the character, line, and global levels. The proposed method achieves a style similarity of 0.751, a content preservation PSNR of 36.9 dB, and an 8%-16% improvement in cross-font generalisation accuracy on unseen characters. For layout generation, the framework maintains a line-spacing fluctuation coefficient of 0.032, achieves a layout aesthetics score of 4.8, and demonstrates strong long-text stability with a cross-page style consistency of 0.94. Ablation studies further confirm the effectiveness of the dynamic weight adjustment mechanism, achieving an optimisation efficiency of 0.98. This work addresses key technical bottlenecks in digital calligraphy generation, providing a practical tool for cultural heritage preservation and a transferable framework for other structured art generation tasks, thereby advancing the integration of artificial intelligence with traditional arts.

Keywords: decoupled representation learning; content-aware; generative adversarial networks; GANs; digital art generation; deep learning.

DOI: 10.1504/IJICT.2026.153802

International Journal of Information and Communication Technology, 2026 Vol.27 No.57, pp.23 - 50

Received: 28 Oct 2025
Accepted: 04 Feb 2026
Published online: 26 May 2026 *

Title: Intelligent generation algorithm for digital image artworks based on decoupling representation and content-aware

Keep up-to-date