视频拷贝检测方法综述

发布时间：2018-08-04 14:58

【摘要】：目前网络上存在着大量的拷贝视频,研究人员长期以来致力于视频拷贝检测技术的研究,特别是近年来随着深度学习方法的引入,又涌现出了一些新颖的检测算法.将对现有代表性的视频拷贝检测方法进行回顾与总结,涵盖视频拷贝检测系统的基本框架与各个主要步骤的不同实现方法,包含视频拷贝检测中的特征提取、建立索引、特征匹配与时间对齐等不同模块.总结的关键技术包括了最新的深度学习方法在其中的应用与取得的突破,主要体现在深度卷积神经网络和双胞胎卷积神经网络方法的应用.此外,还将详细介绍目前常用的5个用于视频拷贝检测评测的数据集及通用的评价标准,并讨论分析一些代表性方法的性能表现.最后,对视频拷贝检测技术未来发展趋势进行展望.
[Abstract]:At present, there are a large number of copies of video on the network. Researchers have been working on video copy detection for a long time. Especially, with the introduction of deep learning methods in recent years, some novel detection algorithms have emerged. This paper reviews and summarizes the existing representative video copy detection methods, including the basic framework of the video copy detection system and the different implementation methods of the main steps, including feature extraction and indexing in the video copy detection. Feature matching and time alignment are different modules. The key technologies include the application and breakthrough of the latest deep learning methods, which are mainly reflected in the application of the deep convolution neural network and the twin convolution neural network. In addition, five commonly used data sets and general evaluation criteria for video copy detection and evaluation are introduced in detail, and the performance of some representative methods is discussed. Finally, the future development trend of video copy detection technology is prospected.
【作者单位】：复旦大学计算机科学技术学院;
【基金】：国家自然科学基金优秀青年科学基金项目(61622204)~~
【分类号】：TP391.41

【相似文献】