Where is the watermark? A block-level interpretable image watermark detection method
Where is the Watermark? Interpretable Watermark Detection at the Block Level
AI summary
A post-hoc image watermarking method combines local embedding with region-level interpretability, and can generate detection maps indicating which regions of an image may contain a watermark or have been modified. The method embeds the watermark signal in the discrete wavelet transform domain using a statistical block strategy, remains robust to common image transformations such as cropping up to half of the image, is sensitive to semantic tampering, and the watermark is highly imperceptible. Compared with existing post-hoc watermarking methods, its detection is more interpretable, and its robustness remains competitive.