哈尔滨商业大学学报（自然科学版）2025，Vol.41Issue(1)：31-37,7.
基于残差对抗神经网络的图像艺术风格迁移
Image art style transfer based on residual antagonistic neural network
杨天 1庾晨龙1
作者信息
- 1. 安徽理工大学 计算机科学与工程学院,安徽 淮南 232001
- 折叠
摘要
Abstract
The artistic style transfer of an image aimed at transforming its style characteristics into the style characteristics of another image on the premise of keeping the content characteristics of the image unchanged.However,the current style transfer methods were all pixel-level convolution in network processing,there was often a general loss of content information and style information,which lead to the leakage of the original content of the generated stylized pictures.In order to solve this problem,this paper proposed a new residual countermeasure network model named RLGAN(Residually Generative Network).The model adopt the structure of multi-layer residual link,which linked the content information in front of the network with the deep features to effectively deal with the problem of information loss.While transmitting style information,RLGAN cleverly protected the content texture structure of the image,so it could generate content-faithful stylized results for a given content map and another style map.In the process of model training,a discriminator and a newly designed loss function were introduced to optimize the updating of decoder parameters and improve the model performance.RLGAN model can generate images with both the texture structure of the original content map and artistic style.This innovative method aimed to overcome the common problem of information loss in traditional style transfer and provided more realistic and artistic image generation.关键词
残差链接/对抗神经网络/结构保护/内容保真/信息损失/样式迁移Key words
residual link/confronting neural network/structural protection/content fidelity/information loss/style migration分类
信息技术与安全科学引用本文复制引用
杨天,庾晨龙..基于残差对抗神经网络的图像艺术风格迁移[J].哈尔滨商业大学学报（自然科学版）,2025,41(1):31-37,7.
