华南理工大学学报(自然科学版) ›› 2012, Vol. 40 ›› Issue (4): 16-22.

• 计算机科学与技术 • 上一篇    下一篇

具有双重检测机制的图像篡改检测算法

胡永健 张尚凡 刘琲贝 谭莉玲   

  1. 华南理工大学 电子与信息学院,广东 广州 510640
  • 收稿日期:2011-10-11 修回日期:2012-01-30 出版日期:2012-04-25 发布日期:2012-03-01
  • 通信作者: 胡永健(1962-) ,男,教授,博士生导师,主要从事多媒体信息安全、图像处理以及模式识别研究. E-mail:eeyjhu@scut.edu.cn
  • 作者简介:胡永健(1962-) ,男,教授,博士生导师,主要从事多媒体信息安全、图像处理以及模式识别研究.
  • 基金资助:

    华南理工大学中央高校基本科研业务费专项资金资助项目( 2012ZM0027)

Image Tampering Detection Algorithm Based on Double Detection Mechanisms

Hu Yong-jian  Zhang Shang-fan  Liu Bei-bei  Tan Li-ling   

  1. School of Electronic and Information Engineering,South China University of Technology,Guangzhou 510640,Guangdong,China
  • Received:2011-10-11 Revised:2012-01-30 Online:2012-04-25 Published:2012-03-01
  • Contact: 胡永健(1962-) ,男,教授,博士生导师,主要从事多媒体信息安全、图像处理以及模式识别研究. E-mail:eeyjhu@scut.edu.cn
  • About author:胡永健(1962-) ,男,教授,博士生导师,主要从事多媒体信息安全、图像处理以及模式识别研究.
  • Supported by:

    华南理工大学中央高校基本科研业务费专项资金资助项目( 2012ZM0027)

摘要: 针对单一基于相关性的图像篡改检测算法难以同时解决虚警和漏检的缺陷,提出一种具有双重检测机制的图像篡改检测算法———第一重检测利用待测图像各区域的相机指纹是否与参考相机的指纹具有相关性来确定疑似篡改图像块; 第二重检测从疑似篡改图像块中抽取特征,送到训练好的支持向量机( SVM) ,由基于图像特征的SVM 对其分类,把误判块和真实篡改块区分开来. 实验结果表明,该算法优于经典的单一基于相机指纹相关性的篡改检测算法.

关键词: 相机指纹, 相关性检测, 模式分类, 篡改检测, 双重检测机制

Abstract:

As the classical correlation-based image tampering detection algorithms cannot avoid false alarm and misdetection,a novel algorithm based on double detection mechanisms is proposed. In this algorithm,first,tampered image blocks under suspicion are picked out according to the correlation between the camera fingerprint in each region of the detected image and that for reference. Then,image features are extracted from each suspicious image block and are sent to a trained SVM ( Support Vector Machine) to construct an image features-based SVM classifier for distinguishing the mis-detected blocks from the real tampered ones. Experimental results demonstrate that the proposed method outperforms the classical one only based on the correlation of camera fingerprints.

Key words: camera fingerprint, correlation detection, pattern classification, tampering detection, double detection mechanism