摘要
动漫手稿上色费时耗力,不断尝试才能确定满意的色彩,一直是动漫行业线稿上色面临的主要问题.文中研究了GAN和Pix2Pix网络模型,采用U-Net结构,优化了基于Pix2Pix算法对灰度动漫手稿进行着色.首先,对灰度手稿数据集进行预处理,使数据集泛化能力增强;其次,对于G网络模型采用8层对称式的U-Net网络,并优化了激活函数,使得G网络模型可以生成更好的训练图像;最后,对于D网络,加入条件y,与G网络不断迭代构建Pix2Pix模型,对数据集anime-faces进行裁剪、归一、填充、色彩通道转换等处理方法,利用深度学习网络模型为灰度动漫手稿自动上色,解决底层信息不变,抑制颜色溢出问题.
Aiming at the time-consuming and labor-intensive coloring of animation manuscripts,it has been a major problem for the animation industry to continuously try to determine the satisfactory color.In this paper,GAN and Pix2Pix network models are studied,U-Net structure is adopted,and the gray animation manuscript is colored based on Pix2Pix algorithm.Firstly,the grayscale manuscript data set is preprocessed to enhance the generalization ability of the data set;secondly,for the G network model,the 8-layer symmetric U-Net network is adopted,and the activation function is optimized,so that the G network model can generate better training images;finally,for the D network,the condition y is added,and the Pix2Pix model is iteratively constructed with the G network.The data set anime-faces are cropped,normalized,filled,and color channel converted.The deep learning network model is used to automatically color the grayscale animation manuscript to solve the problem that the underlying information is unchanged and the color overflow is suppressed.
作者
徐成俊
陈怀圆
XU Cheng-jun;CHEN Huai-yuan(School of Digital Media,Lanzhou University of Arts and Science,Lanzhou 730030,China;School of Electronic and Information Engineering,Lanzhou Jiaotong University,Lanzhou 730070,China)
出处
《兰州文理学院学报(自然科学版)》
2023年第3期63-68,共6页
Journal of Lanzhou University of Arts and Science(Natural Sciences)
基金
甘肃省自然科学基金(22JR5RA217)
兰州市科技计划项目(2022-2-111)。