Anti-Forensics of Image Contrast Enhancement Based on Generative Adversarial Network

Table 2

The average detection accuracies of three CE forensic detectors to four types of anti-forensic images. The lower accuracy indicates the better attack effectiveness of the anti-forensic method. The best attack results are marked in bold.

Attack	Detector					Average
Cao et al. [8]	P-CNN [16]	0.8055	0.6953	0.6120	0.6793	0.6980
	H-CNN [16]	0.0000	0.0000	0.0165	0.0358	0.0131
	GLCM-CNN [24]	0.0013	0.0018	0.0208	0.0410	0.0162
Barni et al. [10]	P-CNN [16]	0.9942	0.9667	0.9725	0.9887	0.9805
	H-CNN [16]	0.5353	0.2025	0.1690	0.2853	0.2980
	GLCM-CNN [24]	0.7620	0.5920	0.5097	0.6657	0.6324
Ravi et al. [11]	P-CNN [16]	0.6378	0.5498	0.6845	0.7075	0.6449
	H-CNN [16]	0.1350	0.0513	0.0658	0.1133	0.0914
	GLCM-CNN [24]	0.0598	0.0335	0.0543	0.0753	0.0557
Our method	P-CNN [16]	0.1704	0.1159	0.0900	0.1452	0.1304
	H-CNN [16]	0.1722	0.0108	0.0262	0.0709	0.0700
	GLCM-CNN [24]	0.0338	0.0179	0.0255	0.0436	0.0302