Researchers at Google have figured out a method for cleanly and automatically removing watermarks on stock photos. The team debuted the method at the 2017 Computer Vision and Pattern Recognition Conference in July.
As most would-be stock photo thieves know, there’s really no easy way to get rid of large, obnoxious watermarks on photos. One could pay for the photo’s rights, or spend an inordinate amount of time in Photoshop painstakingly erasing the watermark — often with unsatisfactory results. Even the latest version of the Adobe software lacks effective tools for the application.
That’s where Google’s new software comes in. It works by scanning a large batch of images with the same overlaid graphics until it detects a pattern and tracks the watermark being used. Once it identifies the watermark, it’s able to automatically remove it — leaving a photo that’s basically indistinguishable from a non-watermarked copy. The drawback is that the software only works on large batches of photos with the same watermark — like, say, from a professional and high-profile stock photo provider.
Of course, Google isn’t exactly in the business of publishing software for intellectual property thieves. In fact, the researchers used this method to develop a way to foil the software they had created.
The team found that randomly deforming or warping the watermark differently on each image renders its software basically ineffective. Even the smallest or most minute geometric variation between watermarks makes it impossible for software to remove the watermark without leaving traces of it behind. Presumably, it’ll make it hard for other non-Google software to remove watermarks, too.
Google’s research allows for a modified approach that will render watermarks much more effective for professional photographers and stock photo companies alike. Which is a good thing all around, because stealing someone else’s work without credit or payment is never cool.