摘要
针对使用卷积神经网络对中国水墨画风格进行学习的过程进行了探讨。首先,分析了VGG19神经网络模型的框架结构,并探讨了如何使用VGG19模型提取艺术风格,并和普通风景图像融合的过程;然后,在理论的基础上,依据中国水墨画的实际特点,通过实验分析寻找合适的卷积层处理内容图像,以及寻找最优的叠加组合对水墨画特征进行提取,并提出了评价图像质量的可视化准则;最后,通过调整内容图像和风格图像的比例系数,得到了符合预期目标的图像,验证了理论的可行性,提出了新的中国水墨画风格图像的风格提取方法。
This paper discusses the process of Chinese ink-painting style learning using convolution neural network.Firstly,the frame structure of VGG19neural network model is analyzed,and the process of using VGG19model to separate and recombine the content and style of artistic images.Secondly,based on the theory,according to the actual characteristics of Chinese ink painting,theappropriate choice of the convoluted layer to process the content image is found and proved byexperimental results.The optimal combination of convoluted layer to extract the style from Chinese ink painting is also found by experiment,and the criteria for visual evaluation of image quality are proposed.Finally,by adjusting the proportion coefficient of the content image and the style image,the expected combined image can be obtained,which verifies the feasibility of the theory and puts forward a new method for Chinese ink-painting style extraction.
作者
王晨琛
王业琳
葛中芹
储开岳
蔡晶
金建华
陈颖
葛云
WANG Chenchen;WANG Yelin;GE Zhongqin;CHU Kaiyue;CAI Jing;JIN Jianhua;CHEN Ying;GE Yun(School of Electronic Science and Engineering, Nanjing University, Nanjing Jiangsu 210023, China;Jiangsu Province Public Security Bureau Material Identification Center, Nanjing Jiangsu 210031, China;Radiotherapy Department, Nantong Cancer Hospital, Nantong Jiangsu 226361, China)
出处
《图学学报》
CSCD
北大核心
2017年第5期754-759,共6页
Journal of Graphics
基金
2015年江苏省政策引导类计划(产学研合作)-前瞻性联合研究项目(BY2015069-06)
2016年度省重点研发计划-社会发展-临床前沿技术(SBE2016750075)
关键词
卷积神经网络
中国水墨画
艺术风格学习
特征提取
convolutional neural network
Chinese ink-painting
artistic style learning
feature extraction
