Camera model and processor identification based on eigen-regularization and extraction technique

Abstract

In the world of mass media, problems with fraudulent images are frequent and troublesome. As such, image forensics has come a long way to ensure the authenticity and validity of images. However, the advancement in camera technology and image editing software has required more effort in the department of camera identification. Using various tests and comparisons in this project, demosaicing features have been found to be the good choice for camera identification with low error rate and reasonable number of raw features. This project has also discovered that there are unique differences not just between camera manufacturers but also within the same brand, in this case, Canon. It was found that there are some discriminant features between different generations of DIGIC image processor, as well as between models. It was also discovered that the accuracy of processor-based and model-based identification improves with the former and deteriorates with the latter, when multiple copies are being trained.