A team of Google researchers demonstrated the dubious feat using a specially designed algorithm that it showed off at the recent Computer Vision and Pattern Recognition Conference in Hawaii. To do it perfectly, however, the algorithm needs to analyze thousands of images with the very same watermark. In that case, stock photo companies, which watermark millions of photos for image previews, seemed to be at particular risk.
Google’s watermark-erasing software scans multiple images until it identifies a recurring pattern, which it concludes to be the watermark. It’s then able to quickly and automatically remove the mark, leaving an unblemished image identical to the one for sale on the stock site.
Shutterstock’s response
But in recent days one of these stock photo companies, Shutterstock, has fought back by reverse engineering Google’s software, according to the The Next Web.
To confuse the algorithm and therefore make it harder for it to effectively remove watermarks, Shutterstock engineers designed a “watermark randomizer” that adds subtle inconsistencies to its marks, ensuring each one is a little different.
By adding these small changes to the watermarks across the millions of images it offers on its site, Shutterstock has been able to prevent the algorithm from identifying repeating patterns that would enable it to remove those watermarks.
“The shapes vary per image and include contributor names,” Shutterstock CTO Martin Brodbeck told TNW. “By creating a completely different watermark for each image, it makes it hard to truly identify the shape.”
Brodbeck added that “changing the opacity and location of a watermark does not make it more secure; however, changing the geometry does.”
Other stock sites would be wise to follow Shutterstock’s example, though it’s likely to be only a matter of time before a programmer creates another watermark-destroying algorithm that can deal with Shutterstock’s solution. And of course, those determined to remove a watermark can still do it manually on a PC, though it can be time consuming and result in a less-than-perfect image.
The battle to effectively protect images online looks set to continue for some time to come.
