基于条件生成对抗网络的手绘图像检索-计算机辅助设计与图形学学报.PDF

29 12 Vol.29 No.12 2017 12 Journal of Computer-Aided Design Computer Graphics Dec. 2017 基于条件生成对抗网络的手绘图像检索 1) 1) 1) 1) 2) 刘玉杰 , 窦长红 , 赵其鲁 , 李宗民 , 李 华 1) ( 中国石油大学(华东)计算机与通信工程学院 青岛 266580) 2) ( 中国科学院计算技术研究所智能信息处理重点实验室 北京 100190) (liuyujie@) : 传统的手绘图像检索方法将自然图像通过边缘检测算法转换成“类手绘图”, 不能很好地减小自然图像与手 绘图像之间的视觉差异. 针对此问题, 提出一种基于条件生成对抗网络的手绘图像检索方法. 首先训练条件生成对 抗网络, 其中生成器由边缘图至自然图像的映射网络构成; 然后通过生成器将手绘图转换为自然图像, 以消除二者 的视觉差异; 最后使用深度卷积神经网络提取深度特征进行相似度度量, 达到检索的目的. 在基准数据库上进行实 验的结果显示, 该方法的检索精度有明显提高. : 手绘图像检索; 条件生成对抗网络; 编码-解码网络; 卷积神经网络 : TP391.41 DOI: 10.3724/SP.J.1089.2017.16596 Sketch Based Image Retrieval with Conditional Generative Adversarial Network Liu Yujie1), Dou Changhong1), Zhao Qilu1), Li Zongmin1), and Li Hua2) 1) (College of Computer Communication Engineering, China University of Petroleum, Qingdao 266580) 2) (Key Laboratory of Intelligent Information Processing, Institute of Computing Technology, Chinese Academy of Sciences, Beijing 100190) Abstract: Traditional methods on sketch based image retrieval leveraged edge detection algorithms to turn natu- ral images into edge maps, but it can not well decrease the visual diversity between natural images and sketches. For this problem, we propose a novel sketch based image retrieval method based on conditional generative ad- versarial networks. Our method is demonstrated as follows: Firstly, we train the conditional generative adversarial networks, of which the generative network is constituted by an edges-to-photo mapping network; secondly, sketch images are converted to natural images by the generative network; thirdly, we use deep convolution neural net- work to extract the deep featu