基于语义分割的织锦类自适应风格迁移算法

通过风格迁移的数字化手段可以辅助艺术作品的创作，但由于织锦类手工艺品具有颗粒感线条、块状色彩、前后景纹理区别较大等特点，使用现有风格迁移算法得到的输出图像在视觉上难以达到令人满意的效果.针对上述问题，提出了一种基于语义分割的织锦类自适应感知域风格迁移算法，将语义分割任务和基于自适应感受域的风格迁移算法相结合，并提出新的内容损失和风格损失.此外，为了解决生成结果图像后景中纹理不均匀的问题，在内容图像上增加高斯噪声用于平滑训练过程中生成图像的后景纹理.实验结果表明，所提算法在织锦作品风格迁移任务中的表现优于现有算法.

Adaptive Style Transfer Method of Brocade Crafts Based on Semantic Segmentation

Neural style transfer has drawn considerable attention to both academic and art field. However,the existing approaches did not perform so good on brocade style transfer because of its grainy texture and blocky colors which is different from painting. A brocade style transfer approach is proposed that combined semantic segmentation task with adaptive style transfer algorithms by using new content loss and style loss. In addition,in order to solve uneven texture of background in the generated image,Gaussian noise is added to the content image to smoothen background texture during training. It is shown that the proposed approach generates brocade stylization outputs that have high quality as compared with other approaches.