局部图像风格迁移绘制算法  

Image style transfer algorithm with local area control

作　　者：孙冬[1] 时宇 高刘雅 李芳芳 高清维[1] 卢一相[1] SUN Dong;SHI Yu;GAO Liuya;LI Fangfang;GAO Qingwei;LU Yixiang(Electric Engineering and Automation College,Anhui University,Hefei 230601,China)

机构地区：[1]安徽大学电气工程与自动化学院,安徽合肥230601

出　　处：《安徽大学学报（自然科学版）》2022年第3期72-78,共7页Journal of Anhui University(Natural Science Edition)

基　　金：国家自然科学基金资助项目(61402003);安徽省高等学校自然科学基金资助项目(KJ2018A0012);赛尔网络下一代互联网技术创新项目(NGII20180612,NGII20180312,NGII20180624)。

摘　　要：提出了一种基于导向滤波的局部图像风格迁移算法.局部图像风格迁移提取图像中感兴趣的部分进行风格迁移,将其渲染成具有铅笔画、水彩画或水墨画等风格的艺术图像.介绍了一种基于导向滤波提取局部图像(也叫图像抠图)的方法,得到二值掩膜图像,利用VGG16网络提取图像的高级特征,通过分离与重组输入图像的内容特征与风格特征的方式,实现局部目标图像的内容与艺术图像的风格的深度融合,并通过掩膜图像去除边缘粗糙,得到精细的风格化图像,从而获得新的局部风格迁移图像.A local image style migration algorithm based on guided filtering was proposed.The partial image style migration extracted the parts of interest in the image for style migration and rendered it into an artistic image with a style of pencil,watercolor or ink painting.This paper introduced a method based on guided filtering to extract local images(also called image maps),obtained binary mask images,extracted advanced features of images by VGG16 network,and separated and reconstructed the content features and style features of input images.The method realized the deep fusion of the content of the local target image and the style of the art image,and removed the edge roughness through the mask image to obtain a fine stylized image,thereby obtaining a new partial style migration image.

关 键 词：图像抠图 卷积神经网络 导向滤波 图像风格化 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]