基于3D残差网络的视频哈希检索  

Video hash retrieval based on 3D residual network

作　　者：冯凯 崔弘 吴锐 FENG Kai;CUI Hong;WU Rui(Wuhan Research Institute of Posts and Telecommunications,Wuhan 430074,China;Nanjing Fiberhome World Communication Technology Co.,Ltd.,Nanjing 210019,China)

机构地区：[1]武汉邮电科学研究院,湖北武汉430074 [2]南京烽火天地通信科技有限公司,江苏南京210019

出　　处：《电子设计工程》2021年第22期128-133,共6页Electronic Design Engineering

摘　　要：视频检索的目标是在海量的视频库中检索到与目标内容相似的视频,针对于当前视频检索方法获取视频信息能力不强、检索准确率不足的问题,提出了一种基于3D残差网络的视频哈希检索算法。设计了一种端到端的网络模型,使用3D残差网络提升视频特征的提取能力,通过哈希层和损失函数的设计,在降低特征维度的同时提升特征的表达能力。在常用的视频数据集UCF101上进行对比实验,结果证明该方法能在占用极少特征空间的前提下,提升相似视频检索的准确率。The goal of video retrieval is to retrieve videos similar to the target content in a massive video library.To solve the problem that the current video retrieval methods have insufficient ability to obtain video information and the retrieval accuracy is insufficient,a 3D residual network based video hash retrieval algorithm was proposed.An end to end network model was designed,and the 3D residual network was used to improve the feature extraction capability of the video,through the design of the hash layer and the loss function,the feature expression ability was improved while reducing the feature dimension.A comparison experiment is conducted on the commonly used video dataset UCF101,and the result proves that the method can improve the accuracy of similar video retrieval with very little feature space.

关 键 词：深度学习 残差网络 视频检索 哈希特征 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]