【摘要】 针对传统织物花型图案设计效率低、成本高、种类单一等问题，提出了一种高效、智能化的基于流行色的花型图案色彩迁移方法。首先根据流行色趋势和色彩理论，通过当前流行色生成兼顾花型图案设计特征的流行色调色盘。然后利用K-means聚类算法，结合生成的流行色调色盘与不同类型的花型图案，构建流行色花型图案数据集。最后设计了色彩迁移模型，对CycleGAN模型进行网络架构优化，引入了边缘保持损失函数和色彩感知损失函数，优化了色彩迁移模型对纹理细节的捕捉能力和色彩真实性的保持效果；同时在模型训练过程中加入椒盐噪声，进一步提升了模型的稳健性。实验结果表明：生成的流行色调色盘在问卷调查中得票率为67.36%,显著高于传统调色盘；生成的色彩迁移图的结构相似性指数和峰值信噪比分别为29.714和0.911,显著高于其它色彩迁移方法。本文提出的色彩迁移方法为织物花型图案的智能化设计提供了一种高效方案，有助于推动纺织服装产业的数字化转型。更多还原

【Abstract】 Objective In order to solve problems of the inefficiencies, high costs, and limited variety in conventional handmade fabric pattern color designs, which are worsened by the frequent shifts in seasonal fashion colors, an efficient, intelligent color-transfer method for fabric patterns is introduced, based on the latest fashion color trends. By closely following fashion color trends, it aims to keep textile and apparel products visually appealing and appealing to consumers. Ultimately, this approach can minimize waste from color-related obsolescence and help the industry quickly adapt to market changes.Method A comprehensive and trend-responsive color palette was developed by integrating fashion color trends, color theory principles, and the unique color attributes of fabric patterns. Using the K-means clustering algorithm, a dataset of fabric patterns incorporating popular color palettes was established. In order to refine the color transfer model, an edge-preserving loss function and a color perception loss function were introduced. This enhanced the ability of the model to capture intricate texture details while maintaining color authenticity. Additionally, salt-and-pepper noise during the model training phase was incorporated to enhance the overall robustness of the model.Results A well-designed survey with 40 participants was conducted to compare two different color palettes in detail. One palette incorporated complex features of fabric patterns, while the other served as a control without these elements. Participants were asked to select their preferred palette based on various factors such as overall fabric design, pattern complexity, and color harmony. This evaluation, based on a diverse set of 90 test samples, ensured the comprehensiveness of the assessment. The results showed that the palette incorporating fabric pattern features received overwhelming positive feedback, with 2 425 votes(67.36% of the total), significantly outperforming the conventional palette and highlighting the strong appeal of fabric-inspired designs. Furthermore, the proposed color transfer model demonstrated excellent performance metrics. Specifically, it achieved a structural similarity index(s) of 0.911, indicating high visual fidelity between the original and transferred images. Additionally, the peak signal-to-noise ratio(p) reached 29.714, further confirming the model′s ability to maintain image quality during the color transfer process. Compared to other conventional color transfer methods that perform well.It outperformed Neural Preset(p=28.295,s=0.894) and AdaIN(s=0.857), These results validate its strengths in maintaining color authenticity and structural fidelity for floral pattern adaptation.In order to verify the contributions of the introduced salt-and-pepper noise, edge loss, and color perception loss to optimizing the color transfer model, ablation experiments were conducted. Compared to the baseline(p=27.596, s=0.806), introducing color perception loss alone(p=28.621, s=0.892) improved color accuracy by 3.7% in p and 10.7% in s, demonstrating superior sensitivity to subtle hue variations. Edge loss achieved an SSIM of 0.844(4.7% gain), effectively preserving structural details, while salt-and-pepper noise enhanced robustness with a 0.9% p increase(27.835). The integrated model combining all components achieved optimal performance(p=29.714, s=0.911). The results indicated that the introduction of the color perception loss function significantly improved the model′s sensitivity to subtle color changes, resulting in more precise and accurate matching of target fashion colors. Thirdly, the edge-preserving loss function played a crucial role in mitigating edge blurring and detail loss during the color transfer process, ensuring that the resulting patterns maintained clear outlines and rich details. Finally, incorporating salt-and-pepper noise during the training phase enhanced the model’s robustness to fluctuations in image quality, simulating real-world disturbances that may occur during image transmission or storage. This enhancement allowed the model to accurately distinguish between true structures and noise, ultimately generating more stable and visually appealing images.Conclusion The proposed fabric pattern color transfer method based on popular colors significantly improves design efficiency and enriches the diversity of fabric patterns while effectively reducing production costs through intelligent means. The feasibility and effectiveness of this method have been verified through empirical experiments. The proposed color transfer method provides an efficient solution for the intelligent design of fabric patterns, accelerating the digital transformation of the textile and apparel industry. Furthermore, by applying the color transfer technology to the fusion design of popular colors and fabric patterns, it not only meet the market′s demand for diversification and personalization but also further respond to the initiative of sustainable development. This approach helps reduce dependence on new materials, aligns with the principles of a circular economy, and provides a more sustainable solution for the textile industry. Ultimately, the research lays the foundation for achieving textile designs that are both environmentally friendly and aesthetically pleasing, making a positive contribution to promoting a sustainable, innovative, and vibrant fashion industry.更多还原