Compression Robustness of Copy-Move Detection Digital Image Forensic Techniques

Digital image forensic techniques, including copymove forgery detection, need to be sensitive against slightest changes in an image. Therefore, their robustness against lossy compression schemes is a crucial issue with respect to their applicability. The robustness of four different copy-move forgery detection techniques is experimentally assessed considering eight distinct lossy compression schemes, including four ISO/IEC JPEG standards as well as four algorithms originating from video intra-frame coding schemes. Overall, we find that the accuracy of the forensic analysis is reduced for increasing compression strength only for the two investigated model-based techniques, while accuracy is constant for low to moderate compression for the learning-based schemes. Also robustness against introduced noise is low for the former techniques while being excellent for the latter ones. However, the relative performance of the distinct compression schemes is different for the considered copy-move forgery detection algorithms, respectively. There is the tendency of intra-frame based video coding schemes being superior to dedicated ISO still image compression schemes. In any case, distinct compression algorithms do not change the ranking of the forensic algorithms in terms of copy-move forgery detection accuracy in different manner.