欢迎访问《图学学报》 分享到：
图学学报

图学学报

• 数字化设计与制造 • 上一篇    下一篇

基于神经风格迁移网络的民族图案创新方法

  

  1. (贵州大学现代制造技术教育部重点实验室，贵州 贵阳 550025)
  • 出版日期:2020-08-31 发布日期:2020-08-22
  • 基金资助:
    国家自然科学基金项目(51865004)；贵州省科技项目(黔科合基础[2017]1046)；黔科合支撑[2017]2016、黔科合支撑[2019]2010)、黔科
    合基础[2018]1049)；黔教合YJSCXJH[2018]088)

Innovative method of ethnic pattern based on neural style transfer network

  1. (Key Laboratory of Advanced Manufacturing Technology of the Ministry of Education, Guizhou University, Guiyang Guizhou 550025, China)
  • Online:2020-08-31 Published:2020-08-22
  • Supported by:
    National Natural Science Foundation of China (51865004); Science and Technology Projects in Guizhou Province ([2017]1046),
    [2017]2016, [2019]2010), [2018]1049); YJSCXJH[2018]088)

PDF (PC)

81

摘要： 针对传统民族图案符号手工提取效率低，图案组合创新设计困难的问题，以苗族
传统的蜡染和挑花图案创新设计为例提出传统民族图案风格创新模型。该模型将改进型的形状
文法与基于深度学习的神经风格迁移网络相结合，提取并编码民族图案构型框架，通过形状文
法生成大量民族图案构型框架图案，使用风格迁移网络快速提取民族图案中的底层特征，在框
架基础上迁移生成创新民族图案设计方案。实验结果表明，该模型可以在指定框架的基础上生
成全新的民族纹饰图案，相对于直接使用神经网络对空白图像进行风格迁移生成的图案更加有
序。生成图案最终应用于苗族织物图案设计中，验证了该方法和设计流程的可行性。

关键词: 卷积神经网络, 形状文法, 神经风格迁移, 挑花, 蜡染

Abstract: Aiming at the problems of low manual extraction efficiency of traditional ethnic pattern
symbols as well as the difficulty of designing creative graphic combination, this paper proposed an
innovative model of traditional ethnic pattern style by taking the batik and cross-stitch work of the
Miao nationality as an example. This model combined the improved shape grammar with neural style
transfer network based on deep learning, to extract e and encode ethnic pattern configuration frames.
Then, a large number of innovative ethnic patterns was generated through shape grammars, and the
basic features of ethnic patterns was extracted quickly using the style-transfer network, so as to
transfer and generate innovative ethnic pattern designs based on this framework. The experimental
results show that the model can generate brand-new ethnic ornament patterns on the basis of the
specified frame, which are more orderly than those generated directly by means of the neural network
transfer. The generated pattern will finally be applied to the design of the Miao fabric patterns, which
verifies the feasibility of this method and design process.

Key words: convolutional neural network, shape grammar, neural style transfer, cross stitch, batik