Abstract: The study found that the PRNU is unique even among cameras of the same type. The identification of source cameras based on PRNU is possible despite heavy JPEG compression that suppresses high-frequency signals such as PRNU. In order to remove DCT (discrete cosine transform)-block edge artifacts introduced by JPEG compression, the authors propose the simple method of averaging multiple pixels into one macro element. This method proved more effective than previously reported methods. The techniques proposed can readily be applied to video footage as well. The research showed that simple and computationally efficient techniques enabled source-camera identification, especially for the closed-set problem. This involves PRNU extraction with a two-dimensional Gaussian filter, detection by calculating a correlation coefficient, JPEG edge artifact suppression by pixel averaging, and scene content suppression by thresholding. The open-set problem performance, however, must be improved. Future work will involve both improving the performance of the identification scheme used in the current research and extending it to different situations and applications.

@article{erwin2009source,
  author       = {Erwin J. Alles and Zeno J. M H. Geradts and Cor J. Veenman},
  url          = {http://www.ncjrs.gov/App/AbstractDB/AbstractDBDetails.aspx?id=248899},
  journal      = {Journal of Forensic Sciences},
  number       = {3},
  volume       = {54},
  year         = {2009},
  title        = {Source camera identification for heavily {JPEG} compressed low resolution still images},
  pages        = {11},
}