In a move that’s bound to make intellectual property providers such as Getty Images Inc. nervous, Google has developed a new algorithm that can easily remove watermarks from images.

Detailed in a paper presented at the 2017 Conference on Computer Vision and Pattern Recognition last month, the algorithm takes advantage of an error in the watermarks are typically applied to photos.  The algorithm scans thousands of sample images from a single stock-photo provider or photographer and compares every one until it identifies a repeating pattern that eventually reveals how the watermark was applied. Using machine learning, the algorithm then restores the image to its prewatermarked state.

“A fact that has been overlooked… is that watermarks are typically added in a consistent manner to many images,” Google research scientists Tali Dekel and Michael Rubinstein wrote in a blog post. “We show that 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. This can all be done automatically, without any user intervention or prior information about the watermark, and by only observing watermarked image collections publicly available online.”

The paper also proposes ways “to make visible watermarks more robust to such manipulations,” or harder to remove. The process involved introducing inconsistencies when applying the watermark to each image, essentially making sure no two watermarks are identical. That doesn’t apply to simply moving the watermark or making small random changes, but by adding random geometric perturbations when applying the watermark to each image.

However, that may not be future-proof either. “While we cannot guarantee that there will not be a way to break such randomized watermarking schemes in the future, we believe (and our experiments show) that randomization will make watermarked collection attacks fundamentally more difficult,” Dekel and Rubinstein added.

Image: Google