摘 要:为了实现基于深度学习的纺织品图案智能辅助设计系统,将分形图案与深度学习风格迁移算法相结合,利用深度学习框架Tensorflow,首先生成基于Mandelbrot 集合的分形图案,接着从图像转换网络结构、感知损失函数等几个方面改进了已有的神经网络风格迁移算法,对分形图案进行风格化设计,使生成的创意图案更符合纺织品图案边界要求高、保持图案细节完整的特点,更加符合纺织品打印要求。
关键词:深度学习;分形图案;风格迁移;卷积神经网络;纺织品图案
中图分类号:TP391.4 文献标识码:A 文章编号:2096-4706(2020)02-0131-03
Intelligent Aided Design of Textile Patterns Based on Deep Learning
WANG Xiaolong1,WANG Ying2
(1.School of Materials Design and Engineering of Beijing Institute of Fashion Technology,Beijing 100029,China;2.Basic Course Department of Beijing Institute of Fashion Technology,Beijing 100029,China)
Abstract:In order to realize the intelligent aided design system of textile pattern based on deep learning,the fractal pattern is combined with the algorithm of deep learning style transfer. Using Tensorflow,the fractal pattern based on Mandelbrot set is generated firstly. Then,the existing neural network style transfer algorithm is improved from the aspects of image conversion network structure and perceptual loss function,the fractal pattern is stylized to make the creative pattern meet the requirements of textile pattern boundary,keep the pattern details complete and meet the textile printing requirements.
Keywords:deep learning;fractal pattern;style migration;convolution neural network;textile pattern
基金项目:北京科技计划项目(ZXKY03180401);北京服装学院大学生创新创业训练计划项目(NHFZ20190153/004)
参考文献:
[1] 王淑缓,杨旭红. 复平面上分形图的生成及在纺织品上的应用 [J]. 丝绸,2017,54(8):56-61.
[2] ULYANOV D,VEDALDI A,LEMPITSKY V.InstanceNormalization:The Missing Ingredient for Fast Stylization [J/OL].arXiv:1607.08022[cs.CV](2016-07-27).https://arxiv.org/abs/1607.08022.
[3] CHEN L C,PAPANDREOU G,KOKKINOS I,et al.DeepLab:Semantic Image Segmentation with Deep Convolutional Nets,Atrous Convolution,and Fully Connected CRFs [J]. IEEE Transactions on Pattern Analysis & Machine Intelligence,2016,40(4):834-848.
[4] LUAN F,PARIS S,SHECHTMAN E,et al. Deep Photo Style Transfer [C]//2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Honolulu,HI,USA,2017(1):6997-7005.
[5] WEISSTEIN E W. Gram Matrix [EB/OL].(2019-11-01).Wolfram MathWorld. http://mathworld.wolfram.com/GramMatrix.html.
作者简介:
王晓龙(1997-),男,汉族,吉林长春人,本科在读,研究方向:轻化工程;
通讯作者:
王颖(1975-),女,回族,河北张家口人,教师,讲师,硕士研究生,研究方向:计算机应用技术。