Embedding Guided End-to-End Framework for Robust Image Watermarking

Table 6

Comparison with other algorithms against the attack types different from those in the training stage.

Algorithms	Resizing (T = 2)	Medium blur (ω = 3)	Salt and pepper noise (p = 0.2)	Gaussian noise (σ = 1.0)
Luo [37]	27.97	29.48	30.15	29.75
Ding [38]	26.64	28.57	29.31	28.43
Baluja [39]	28.89	30.32	31.67	30.46
Ours	32.87	33.91	34.02	33.52