基于神经网络的手绘服饰图纹上色及风格迁移  

Costume Pattern Sketch Colorization and Style Transfer Based on Neural Network

作　　者：蔡兴泉[1] 李治均 奚梦瑶 孙海燕[1] Cai Xingquan;Li Zhijun;Xi Mengyao;Sun Haiyan(School of Information Science and Technology,North China University of Technology,Beijing 100144,China)

机构地区：[1]北方工业大学信息学院,北京100144

出　　处：《系统仿真学报》2023年第3期604-615,共12页Journal of System Simulation

基　　金：北京市社会科学基金(20YTB011)。

摘　　要：针对图纹上色容易色彩溢出、风格迁移缺少布料纹理特征等问题,提出基于神经网络的手绘服饰图纹上色及风格迁移方法。初始化数据集,收集服饰图纹图像,提取服饰图纹黑白线稿,合成具有颜色特征的手绘图像,构建风格数据集;构建条件生成对抗网络模型,基于该生成器模型实现对具有颜色信息的线稿图像上色;构建卷积神经网络模型,利用该模型计算内容图的内容特征并结合Gram矩阵计算风格图的风格特征,输出令人满意的服饰图纹迁移图像。实验结果表明,该方法生成的图像具有真实的服饰图纹颜色分布,具有较好的布料材质感。Aiming at the problems of color overflow in pattern sketch colorization and lack of fabric texture features in style transfer,this paper proposes a method of costume pattern sketch colorization and style transfer based on neural network.This paper initializes the data set,collects the costume pattern image,extracts the costume pattern sketch,synthesizes the costume pattern sketch with color features and constructs the style data set.The research builds the conditional generative adversarial nets and achieves the costume pattern sketch with color features colorization based on the generator.The study constructs a convolutional neural network model,uses the model to calculate the content features of the content map and uses the Gram matrix to calculate the style features of the style map,introduces weight parameters to optimize the loss function,and outputs a satisfactory costume pattern transfer image.The experimental results show that the generated image has a real costume pattern color distribution and a good sense of fabric material.

关 键 词：手绘服饰图纹 线稿图像上色 风格迁移 条件生成对抗网络 卷积神经网络 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]