Camera Device Identification and the Effects of Underexposure

Abstract

Technology today allows a photograph from a digital camera to be matched with the camera that took it. However, the matching software was created over 10 years ago using data that is not necessarily representative of today’s data. The objective of this work is to look at the case where photographs that are acquired are underexposed. The data used for these experiments are taken from StegoAppDB, a forensic imaging database. We obtain the PRNU fingerprints of the cameras and test against fingerprints from other images. Using the peak to correlation energy (PCE) as a similarity metric, the image fingerprint and camera fingerprint are compared. A PCE score less than 60 signifies that the image did not originate from the camera -- not a match; otherwise, it did. Error rates for auto-exposed and under-exposed images were based on these comparisons. Using a carefully curated data set of smartphone images, a PCE threshold of 60 results in a false positive rate of 5.86*10-3 when comparing autoexposure fingerprints to underexposure test images.