Google has found a way to remove watermarks from pictures without too much of a hassle. By stripping an image of the protections offered by watermarks, it makes them easy to share across the Internet without proper credit being given.
Described in a paper called “On the Effectiveness of Visible Watermarks,” the breakthrough had already been presented at a conference on computer vision in Hawaii last month.
In a blog post from Google, signed by Tali Dekel and Michael Rubinstein, both research scientists at the company, they explain that they’re treating the matter just as any vulnerability they might find out there. Not only do they detail how they reached this discovery, but they also make several proposals about how to help the photography and stock image communities adapt to the future, and how the members of these communities can protect their copyrighted content and creations.
The two researchers point out that the main problem with the situation is that watermarks are predictable. They explain this that watermarks often look the same, are placed in the same positions and have the same size. “This consistency can be used to invert the watermarking process – that is, estimate the watermark image and its opacity, and recover the original, watermark-free image underneath,” the blog reads.
This, they say, can be done automatically, without any user intervention, by simply observing watermarked image collections that are currently available online.
The Google team built a software and trained it to identify watermark patterns. Then, when they had fed enough watermarked images in the system, they managed to separate the watermark’s components from the rest of the image. Once this first hurdle was passed, the algorithm could remove the watermarks from any photo.
So what can photographers to do to protect their work? Well, they can make some changes to the watermarks they used. Once there’s a change in how it looks, such as text or logo warping, the algorithm no longer knows how to properly remove the watermark, leaving behind notable traces.
“In a nutshell, the reason this works is because removing the randomly warped watermark form any single image requires to additionally estimate the warp field that was applied to the watermark for that image – a task that is inherently more difficult. Therefore, even if the watermark pattern can be estimated in the presence of these random perturbations (which by itself is nontrivial) accurately removing it without any visible artifact is far more challenging,” they say.
While this seems like the way to go right now, there’s nothing to say that a more advanced algorithm won’t be created in the future to bypass even these protections.