Performance and Robustness Analysis for Some Re-sampling Detection Techniques in Digital Images

Abstract

In order to create convincing forged images, manipulated images are usually exposed to some geometric operations which require a re-sampling step. Therefore, detecting traces of re-sampling became an important approach in the field of image forensics. There are many re-sampling detection techniques described in the literature, but their performance has been often examined with a small dataset. Besides, performance and robustness have been tested under different conditions, so it is difficult to evaluate and compare them. In this paper, we analyze the performance of some selected re-sampling detection techniques by using a common testing framework and a large dataset. We also employ several kinds of image post-processing to defeat the detectors in order to evaluate their robustness and security. We show that the tested techniques obtain the best results when detecting up-sampled images. Unfortunately, most techniques are not secure and they can be defeated on different levels by applying post-processing operations to the forged images.