Is watermarking still a sound way to protect your images? And what threats do AI-based tools pose?
Digital image watermarking is one of the most popular ways to protect images posted online.
Watermarks are easy to apply to images of all varieties, and they give the photographer or copyright owner the chance to promote themselves or their brand beyond their existing audience.
But just how secure is digital image watermarking today? And how can those with assets to protect safeguard their work from potential threats?
What is a watermark and how does it work?
The idea behind watermarking is simple. By including a semi-transparent line of text, symbol or logo into an image, you make it clear that the rights to it belong to a specific photographer or company, and that it cannot simply be used without authorization.
Watermarks are usually embedded so that they become part of the image itself. This means that if the image does get stolen in some way, it will remain in place.
The watermark itself often identifies the photographer or copyright holder, although this isn’t always the case. The image may, for example, simply be marked with a copyright symbol, or with a more elaborate design or repeating pattern that makes use of multiple lines or copyright symbols (or both) to provide greater protection.
However it appears, an effective watermark serves a number of purposes. First, by making the presence of copyright clear, it underlines that the image in question cannot simply be used without the copyright holder’s consent, which, in turn, serves as a deterrent for would-be thieves.
It also limits the likelihood of it being used, were it to be stolen. Furthermore, in addition to its marketing purposes, it also makes it far easier to prove an image’s true ownership in the event of any unauthorized use, assuming the watermark itself identifies the photographer or copyright holder.
Where things can go wrong
The effectiveness of a watermark depends on many factors. Its visibility, for example; a watermark that’s barely visible is likely to be a less effective deterrent than one that can be easily seen. A watermark that isn’t too visible may also be easier to remove than a more prominent one, in that efforts to lift it away are less likely to adversely affect that image underneath it.
Placement is also key. A watermark that’s placed in a relatively featureless area may be more easily removed that one that’s placed over more complex details. A thief is more likely to attempt to lift a watermark from an image if they believe they can keep the image itself relatively intact.
The rise of AI-based solutions
Perhaps the greatest threat to the watermark in recent times has been the rise of sophisticated software algorithms that can be used to remove watermarks from images.
While we’ve had similar tools incorporated into popular image editing programs for some time, solutions designed specifically to remove watermarks also exist in the form of standalone packages, and many of these are available for free. Furthermore, as they are increasingly making use of artificial intelligence (AI) to better detect and remove watermarks, they now typically require less human intervention than they used to.
Such tools have clear appeal to those looking to lift watermarks from individual images, but it’s perhaps only a matter of time before more automated tools for batch watermark removal become available.
In 2017, researchers at Google released a paper titled On the Effectiveness of Visible Watermarks. The paper detailed a watermark-removing algorithm developed by the researchers, and highlighted the ease with which this could be used to exploit the consistent manner in which watermarks tend to be applied to multiple images.
The team of researchers found that their algorithm continued to work when the watermark was placed in different areas within a series of images, but also noted that it was less effective when there were spatial or geometric discrepancies between watermarks.
Changing watermarks between images obviously creates more work for the copyright owner, in that a series of images cannot be watermarked in one action. Those taking the effort to protect their work, however, may consider this to be a price worth paying for the additional layer of security. Indeed, the paper argues this point, stating that “visible watermarks should be designed to not only be robust against removal from single images, but to be resistant against removal from image collections as well.”
Google isn’t the only major company to have worked on this type of tool. In 2018, Nvidia announced that its research had led it to develop an AI algorithm to effectively lift away artifacts from images, such as noise and text. Its effectiveness, which you can see in the video below, will no doubt alarm many photographers.
The fact that these tools are not publically available – at least not yet – will be of some comfort to photographers. That said, if we assume that an effective algorithm is made commercially available in some form in the future, knowing at least some of its weaknesses is undoubtedly useful for watermarks created today.
While tools exist to help photographers find stolen images online, when you consider that watermarked images may well be used offline as well as online, keeping track of them even more difficult. One particularly egregious example of this was the case of a hand mixer spotted in a Polish supermarket, whose packaging included images with a very obvious Shutterstock watermark. This is admittedly an extreme case, and may well be down to human error rather than anything more malicious, but this is just one of countless examples where, for whatever reason, a watermarked image has appeared somewhere it shouldn’t have.
What options do today’s photographers have?
Photographers worried about increasingly sophisticated watermark-removal technology have a number of options.
One approach is to simply increase the size and prominence of a watermark. This would conceal more of the image and would – at least in theory – make it more difficult to remove without severely affecting the image it protects. Such a move divides photographers; some prioritize security over presentation and would see it as a worthy approach, while others would consider this to be too detrimental to the viewing of the image. Yet, as the tools from Google and Nvidia show, even very prominent watermarks may only provide so much protection.
Another approach tackles the ease with which images can be downloaded or scraped by bots from websites and combines this with a dynamic form of watermarking. Here, images are streamed rather than embedded in websites, which strips the downloadable JPEG from the website’s source code and prevents right-clicking or drag-and-drop saving. Meanwhile, the dynamic watermarking component provides the copyright holder with constant control over how the watermark appears, so that it can be changed as and where necessary for many images at once.
• Watermarks serve a number of purposes, from deterring thieves to giving photographers a chance to market their brand to new audiences.
• The design and placement of a watermark is important to consider, as this will affect how successfully it can be manually removed.
• The tools we have at our disposal are far less sophisticated than those developed by researchers at the likes of Google and Nvidia in the past few years.
• More robust ways of protecting images look at the issue more holistically, taking the ways in which images are consumed and typically stolen into account.