![]() In pixels where there isn't a watermark, nothing happens since they're identical. To do this, you combine the pixels of two watermarked images in imagemagick or the like (take the "highest" of the two, the "lowest" of the two, the average of the two, etc). Once you have multiple releases of the same image with different watermarks applied (which they're doing per-sale), it leaks information like a sieve because now all an attacker needs to do is run a comparison between different watermarks of the same image to detect the watermark information.īut our task is even simpler, because we don't want to extract the data, just destroy it. It's quite clever, but only when there's ONE watermarked image per original (and the original is never leaked). Their algorithm is designed around regularly spaced regions that are adjusted in the chroma space to give any image a set of "bits" that remain detectable, even after image resizing, converting to greyscale, etc when compared to the original. Imatag has an Achilles heel because it's not actually designed to identify leakers, only copyright infringers.
0 Comments
Leave a Reply. |