Read the full-text of this article for free

Title: Image retrieval of hand-drawn sketches in Shu embroidery pattern based on CycleGAN and triplet network

Authors: Ran Jia; Huifang Ma

Addresses: College of Art, Beijing Union University, Beijing, 100101, China ' College of Art, Beijing Union University, Beijing, 100101, China

Abstract: To unify the style of Shu embroidery pattern hand drawn sketches and embroidery images, and improve the efficiency and robustness of Shu embroidery pattern hand-drawn sketch image retrieval, this study proposes a Shu embroidery pattern hand-drawn sketch image retrieval method based on CycleGAN and triplet network. A closed-loop framework was developed, including input, feature extraction, cross-domain transformation between sketch and embroidery images using CycleGAN, optimisation of feature representation using a triplet network, and ultimately retrieval output. The experiment showed that the area under the 'false positive rate - true positive rate' curve of the research method was about 97.2%. When the training epochs were 80, the learning perception image block similarity value of the research method fully converged and reached a stable value of 0.01. The above results indicate that the research method has good retrieval efficiency, accuracy, and generalisation.

Keywords: cycle generative adversarial network; triple network; text retrieval; hand-drawn sketch retrieval; image cross-domain conversion; cross modal retrieval.

DOI: 10.1504/IJICT.2026.153266

International Journal of Information and Communication Technology, 2026 Vol.27 No.37, pp.1 - 28

Received: 16 Oct 2025
Accepted: 09 Dec 2025
Published online: 29 Apr 2026 *