信息取证中图像重采样检测  

Re-sampling detection in information forensics

作　　者：朱秀明[1] 宣国荣[1] 姚秋明[1] 童学锋[1] 施云庆 

机构地区：[1]同济大学计算机科学与技术系,上海200092 [2]新泽西理工学院电气与计算机工程系,美国新泽西nj07102

出　　处：《计算机应用》2006年第11期2596-2597,2601,共3页journal of Computer Applications

基　　金：国家自然科学基金资助项目(90304017)

摘　　要：经过重采样后的数字图像,实质是原始图像信号和周期信号的叠加。通过期望最大化算法对两种信号进行聚类分析,便可以检测其中的重采样痕迹。通过对已有的模型进行改进,将原始信号分布代替均匀分布并增加了先验概率的迭代,使得分布情况更符合实际。同时,通过增加小补偿量的方法避免EM算法可能遇到的奇异点。实验证实该算法可以检测出对于未压缩图像的缩小、放大和旋转及图像是否被JPEG压缩过。A digital image generated by re-sampling is the combination of the original image signal and periodical signal. Through the analysis of the two original signals by using EM algorithm, the traces of re-sampling could be detected. Through the modification of the present model, the uniform distribution was supplanted by the original image distribution and a priori probability updating was added in order to approach the realistic distribution. The possible singular points during the EM iterations were avoided by adding a small threshold. Experiment results demonstrate the effectiveness of the modified model in detecting image re-sampling. Furthermore, it can detect re-sampling traces on uncompressed bmp or tiff images, and it even can detect whether an image is JPEG compressed accurately.

关 键 词：期望最大化算法 数字取证 重采样 参数估计 

分 类 号：TP309.2[自动化与计算机技术—计算机系统结构]