网络与信息安全学报 ›› 2019, Vol. 5 ›› Issue (5): 64-79.doi: 10.11959/j.issn.2096-109x.2019052

• 学术论文 • 上一篇    下一篇

融合空间约束和梯度结构信息的视频篡改检测算法

普菡1,2,3, 黄添强1,2,3(), 翁彬1,2,3, 肖辉1,2,3, 黄维1,2,3   

  1. 1 福建师范大学数学与信息学院,福建 福州 350007
    2 福建省大数据挖掘与应用工程技术研究中心,福建 福州 350007
    3 数字福建大数据安全技术研究所,福建 福州 350007
  • 修回日期:2019-06-06 出版日期:2019-10-15 发布日期:2019-11-02
  • 作者简介:普菡(1995- ),女,河南平舆人,福建师范大学硕士生,主要研究方向为信息安全、数字多媒体取证。|黄添强(1971- ),男,福建仙游人,博士,福建师范大学教授,主要研究方向为机器学习、数字多媒体取证。|翁彬(1981- ),男,福建福州人,博士,福建师范大学讲师,主要研究方向为机器学习及应用。|肖辉(1991- ),男,福建建瓯人,福建师范大学硕士生,主要研究方向为信息安全、数字多媒体取证。|黄维(1994- ),女,福建莆田人,福建师范大学硕士生,主要研究方向为信息安全、数字多媒体取证。
  • 基金资助:
    国家重点研发计划专项基金资助项目(2018YFC1505805);应用数学福建省高校重点实验室基金资助项目(SX201803)

Video tampering detection algorithm based on spatial constraint and gradient structure information

Han PU1,2,3, Tianqiang HUANG1,2,3(), Bin WENG1,2,3, Hui XIAO1,2,3, Wei HUANG1,2,3   

  1. 1 Mathematics and Informatics,Fujian Normal University,Fuzhou 350007,China
    2 Fujian Provincial Engineering Research Center of Big Data Analysis and Application,Fuzhou 350007,China
    3 Digital Fujian Big Data Security Technology Institute,Fuzhou 350007,China
  • Revised:2019-06-06 Online:2019-10-15 Published:2019-11-02
  • Supported by:
    National Key Program for Developing Basic Science(2018YFC1505805);Applied Mathematics Fujian Provincial Key Laboratory Project(SX201803)

摘要:

相邻帧间相似性原理的传统视频被动取证方法会对画面运动剧烈的视频发生大量误检测,针对这个问题,提出了一种融合空间约束和梯度结构信息的视频篡改检测方法。首先,利用空间约束准则,提取低运动区域和高纹理区域,并将两个区域进行融合,获取顽健的量化相关性丰富区域用于提取视频最优相似性特征;然后,改进原有特征的提取和描述方法,运用符合人类视觉系统特性的梯度结构相似性 GSSIM 来计算空间约束相关性值,最后,利用切比雪夫不等式对篡改点进行定位。实验证明,针对画面运动剧烈的视频,所提算法误检率更低,精确度更高。

关键词: 空间约束, 量化相关性丰富区域, 梯度结构相似性, 画面运动剧烈的视频

Abstract:

The traditional video passive forensics method using only the principle of similarity between adjacent frames will cause a lot of false detection for the video with severe motion.Aiming at this problem,a video tamper detection method combining spatial constraints and gradient structure information was proposed.Firstly,the low motion region and the high texture region were extracted by using spatial constraint criteria.The two regions were merged to obtain the robust quantitative correlation rich regions for extracting video optimal similarity features.Then improving the extraction and description methods of the original features,and using the similarity of the gradient structure in accordance with the characteristics of the human visual system to calculate the spatial constraint correlation value.Finally,the tampering points were located by the Chebyshev inequality.Experiments show that the proposed algorithm has lower false detection rate and higher accuracy.

Key words: spatial constraints, the quantitative correlation rich regions, GSSIM(gradient structure similarity), videos with severe motion

中图分类号: 

No Suggested Reading articles found!