Google researchers have described that computer algorithms can easily remove a watermark from a photo and leave an image without copyright protection. The company described this in detail in a scientific paper entitled "On the Efficiency of Visible Watermarks" .
"As often happens with vulnerabilities found in operating systems, applications or protocols, we want to unleash this vulnerability and offer solutions to help the communities of stock images and photos adapt and better Protect their author's content and creations, " – wrote Google researchers Tali Dekel and Michael Rubinstein
According to them, the main problem of the existing process of imposing watermarks is a high level of coincidence of styles. "We demonstrate that this similarity can be used to invert the process of imposing watermarks – that is, evaluating the image of the watermark and its opacity and restoring the underlying image underlying it, – the researchers said. – All this can be done automatically, without user intervention and receiving additional information about the watermark – solely through the study of watermark image collections available on the Internet publicly. "
The team was able to teach the software to isolate models of applying watermarks and using a special technology to separate their components from the main picture. The algorithm can remove a certain character or even look like it from any image, because it already knows its parameters like transparency and structure.
As a measure of protection, researchers propose to add elements of randomness to watermarks. This, in particular, can be an accidental change in the shape of the applied image. Changes must leave visible artifacts after the deletion of the mark.