Abstract: Detecting the (brands and) models of digital cameras from given digital images has become a popular research topic in the field of digital forensics. As most of images are JPEG compressed before they are output from cameras, we propose to use an effective image statistical model to characterize the difference JPEG 2-D arrays of Y and Cb components from the JPEG images taken by various camera models. Specifically, the transition probability matrices derived from four different directional Markov processes applied to the image difference JPEG 2-D arrays are used to identify statistical difference caused by image formation pipelines inside different camera models. All elements of the transition probability matrices, after a thresholding technique, are directly used as features for classification purpose. Multi-class support vector machines (SVM) are used as the classification tool. The effectiveness of our proposed statistical model is demonstrated by large-scale experimental results.

@inproceedings{Xu:2009aa,
  publisher    = {Springer Verlag},
  author       = {Guanshuo Xu and Shang Gao and Yun Qing Shi and RuiMin Hu and Wei Su},
  url          = {http://www.springerlink.com/content/x200104646n09w21/},
  series       = {Lecture Notes in Computer Science},
  booktitle    = {Digital Watermarking. 8th International Workshop, IWDW 2009, Guildford, UK, August 24--26, 2009, Proceedings},
  volume       = {LNCS 5703},
  year         = {2009},
  editor       = {Anthony T.S. Ho and Yun Q. Shi and H.J. Kim and Mauro Barni},
  address      = {Berlin, Heidelberg},
  title        = {Camera-model identification using {M}arkovian transition probability matrix},
  pages        = {294--307},
}