基于深度神经网络的图像风格迁移系统开发  

Development of image style migration system based on depth neural network

作　　者：黄凯怡 姜佳琪 方建文 杨彩云 HUANG Kaiyi;JIANG Jiaqi;FANG Jianwen;YANG Caiyun(College of Electrical and Information Engineering,Quzhou University,Quzhou,Zhejiang 324000,China)

机构地区：[1]衢州学院电气与信息工程学院,浙江衢州324000

出　　处：《计算机应用文摘》2023年第15期56-59,共4页Chinese Journal of Computer Application

基　　金：衢州市科技计划项目(2020K21);国家级大学生科技创新项目(202111488005,202111488059);衢州学院横向项目(H2021321)。

摘　　要：传统的图像风格迁移程序通过人工建模只能局限于单一风格迁移,且资源消耗过大。2015年Gatys等提出了一种基于卷积神经网络的图像风格迁移算法,通过这一种算法便能生成各种具有原内容和新风格的合成图像。文章利用深度学习算法,对卷积神经网络模型和风格迁移算法进行图像风格特征提取的研究,在MXNET深度计算引擎上开发图像风格迁移系统(Styler),将一副图像的风格迁移到另一副图像,其成果可应用到设计、娱乐和社交等领域。Traditional image style transfer programs can only be limited to a single style transfer through artificial modeling,and consume too much resources.In 2o15,Gatys et al.proposed an image style transfer algorithm based on convolutional neural networks,through which various composite images with original content and new styles can be generated.This paper uses deep learning algorithm to study the image style feature extraction of convolutional neural networks model and style transfer algorithm,and develops an image style transfer system(Styler)on the MXNET deep computing engine to transfer the style of one image to another.Image,its results can be applied to design,entertainment and social fields.

关 键 词：深度神经网络 图像风格迁移 卷积神经网络 VGG-19模型 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]