摘要
在社交网络分享平台上,照片艺术风格处理成为热点应用。运用卷积神经网络方法将图像内容与风格进行分离,生成对应的图像表述;选择原有图像内容与艺术作品的图像风格,采用Keras框架和计算数据第三方库,计算得到内容与风格损失的初始总损失;通过不断地优化与迭代,尽量降低总损失数值,根据这些数值重新糅合生成新的图像实现图像风格迁移。测试数据采用学校风景图片与梵高油画作品实现风格迁移生成艺术化图像。
On the social network sharing platform,photo art style processing has become a hot application.Uses the convolutional neural network method to separate the image content from the style to generate the corresponding image representation;selects the original image content and the image style of the artwork,and uses the Keras framework and the third-party library of computing data to calculate the initial con-tent and style loss.Total loss;by continuously optimizing and iterating,the total loss value is minimized,and finally,the new image is re-combined based on these values to achieve image style migration.The test data uses the landscape pictures of the school and the Van Gogh oil paintings to realize the style migration to generate artistic images.
作者
窦亚玲
周武彬
季人煌
DOU Ya-ling;ZHOU Wu-bin;JI Ren-huang(College of Information Science and Engineering,Hunan Normal University,Changsha 410081)
出处
《现代计算机》
2018年第20期47-51,60,共6页
Modern Computer
基金
湖南省教育厅教改项目
湖南省教育厅科研项目(No.14C0681)
关键词
卷积神经网络
风格迁移
纹理合成
Convolutional Neural Network
Style Migration
Texture Synthesis
