摘要
提出一种基于卷积神经网络的翻拍图片哈希检索方法,基本思路是:在卷积神经网络中加入哈希层,提取哈希层输出的二值哈希码用于图像检索;检索匹配阶段采取由粗到精的分级检索策略,依次计算查询图像与图片库中图像之间的距离,得到最终的检索结果。在翻拍广告图片集上的实验结果表明:该方法对翻拍图片旋转、缩放、遮挡、反光等不敏感,与仅使用fc7层输出进行检索相比,平均检索精度值增加了12.1%。
This paper proposes a hash retrieval method for recaptured images based on convolutional neural network.The basic idea is to add a hash layer to the convolutional neural network,and extract the binary hash code output by the hash layer for image retrieval.Then,a hierarchical search strategy from coarse-to-fine is used to calculate the distance between the query image and the image in the image dataset in order to obtain the final search result.The experimental results on the recaptured advertisement image collection show that the method in this paper is insensitive to recaptured image rotation,scaling,occlusion,reflectionetc.Compared with only the output of fc7 layer for retrieval,the mean average precision increased by 12.1%.
作者
李晶
王翾
刘守训
LI Jing;WANG Xuan;LIU Shou-xun(School of Information and Communication Engineering,Communication University of China,Beijing 100024,China)
出处
《中国传媒大学学报（自然科学版）》
2020年第2期65-71,共7页
Journal of Communication University of China：Science and Technology
关键词
图像检索
卷积神经网络
翻拍图片检索
分级检索
image retrieval
convolutional neural network
recaptured image retrieval
hierarchical retrieval
