Digital Camera Identification

In this work, we focus on blind identification of digital cameras used in cellular phones. Since cellular phones are so widely deployed, we thus indirectly bring a solution to their authentication. Digital cameras possess a chain of processing operations such as signal quantization, white balance, demosaicking, color correction, gamma correction, filtering and JPEG compression. Since these algorithms or their parameter settings are proprietory, they leave characteristic footprints. We pick up these footprints as camera brand or make signatures and develop a recognition algorithm. We show that up to a dozen cameras can be reliably identified using the statistics of the signature signals, even when the images are subjected to manipulations. The performance improves with fusion schemes, applied at feature level, decision level and score level.