基于插值检测的扫描图像源辨识方法  

Scanned Image Source Identification Based on Interpolation Detection

作　　者：房穗韬[1] 方艳梅[1] 

机构地区：[1]中山大学信息科学与技术学院,广东广州510006

出　　处：《中山大学学报（自然科学版）》2013年第5期19-25,共7页Acta Scientiarum Naturalium Universitatis Sunyatseni

基　　金：国家973计划资助项目(2011CB302204);国家科技支撑计划资助项目(2012BAK16B00)

摘　　要：针对扫描图像真实性的问题,提出了应用EM算法与扫描仪传感器噪声的自然统计特性相结合的识别方法。该识别方法一共生成63维特征,用SVM进行训练,通过大量实验图像,文中对取证方法效果进行了分析研究。结果显示,扫描图像虽然在生成过程中不存在CFA插值过程,但是在被显示前,仍经过了插值运算,傅立叶频谱图仍具有周期性。该算法对SG图像(扫描图像)与CG图像、SG图像与DG图像(数码相机生成图像)之间的识别,达到了较理想的效果。Abstract： Since scanned images are not generated by CFA process, a part of researchers have studied scanned image forensic around sensor pattern noise. A method is proposed to contain the statistical fea- tures of the noise image and features derived from EM algorithm for the authenticity of the scanned image. This method generates a total of 63 dimensional feature classified by SVM. By analyzing the evidence ef- fectives and testing a large number of images, the work results show that although scanned images is not generated with CFA process, scanned images also has experienced interpolation on its Fourier spectrum map. Moreover, for the purpose of distinguish between scanned images and other images （ CG, DG） , this method achieves the desired results.

关 键 词：图像取证 扫描图像 传感器噪声 插值检测 EM算法 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]