Countering median filtering anti-forensics and performance evaluation of forensics against intentional attacks

摘要

Median filtering forensics and its anti-forensic attack have received considerable attention since median filtering can be used for both image enhancement and anti-forensic purposes. A median filtering anti-forensic attack method by adding uniformly distributed noise was proposed in an image pixel domain. However, we observe that this attack method leaves visible traces in the histogram of its median filtering residual (MFR) and can be detected using a histogram bin ratio of its MFR in the textured area. In order to eliminate this trace left in the MFR, we propose to adding noise adaptively in pixel domain to keep a constant minimal SNR. The performance of several forensic methods are evaluated under several attacks, it shows that the AR (autoregressive) forensic method has the most robustness against intentional attacks compared with the other forensic methods.