基于ChipGAN-ViT模型的汉绣艺术风格迁移与模拟

Transfer and Simulation of Chinese Embroidery Art Style Based on ChipGAN-ViT Model

针对真实图像与汉绣图像在风格迁移融合过程中产生的针法工艺模糊和边界伪影问题,提出了基于ChipGAN-ViT模型的汉绣风格迁移方法。由于刺绣纹样内部与背景留白具有不同的线迹填充效果,算法首先利用ChipGAN-ViT模型对前景进行纹理重构,再利用循环生成对抗网络对风格图像和内容图像进行风格迁移;其次,采用Sobel算子对汉绣图像进行边缘轮廓提取,以满足汉绣数字化模拟的内容图像需求;最后,对生成的风格迁移图像进行超分辨率处理获得最终汉绣数字化图像。实验结果表明:该方法可有效模拟出汉绣平顺且配色丰富的艺术特点,相比传统的ChipGAN、CNN算法迁移时间缩减了30.58%和41.52%。所提出的汉绣风格迁移方法是对风格迁移技术的有效补充,为汉绣图案的创新设计提供了新的可能。

To address the issues of ambiguity in the stitch process and boundary artifacts during style transfer and fusion between real images and Han embroidery images,this paper proposes a method for Han embroidery style transfer based on the ChipGAN-ViT model.Firstly,the algorithm utilizes the ChipGAN-ViT model to reconstruct the texture of the foreground,taking into consideration that different stitch filling effects exist between the interior of embroidery patterns and blank backgrounds.Then,a cyclic generation adversarial network is employed to transfer both style and content images.Additionally,Sobel operator is utilized to extract edge contours from Han embroidery images in order to meet content image requirements for digital simulation of Han embroidery.Finally,super resolution processing is applied to enhance details in the style transferred image,resulting in a final digital representation of Han embroidery.Experimental results demonstrate that this approach effectively simulates smoothness and vibrant artistic features of Han embroidery while reducing migration time by 30.58%compared with traditional ChipGAN algorithms and 41.52%compared with CNN algorithms respectively.This method serves as an effective supplement to style transfer technology while offering new possibilities for innovative design of Han embroidery patterns.