Image Detection: Watermarking vs. Fingerprinting


The image detection field is changing rapidly. For years I’ve been predicting the growth in image searching and it appears that is finally coming to be. Not only has PicScout has begun public testing on its newest offerings and other announcements are in the works.

As such, over the next few weeks and months, it seems almost certain that I’ll be posting articles about new technologies and advancements in this field as it continues to heat up and prices begin to drop. This means that many photographers, including smaller ones, will be confronted with the question about how to best protect their images and track their usage.

Three years ago, this was almost unthinkable. Limited options and high prices made such tracking attainable only to large companies but now we’re preparing to enter a very different age for visual artists when it comes to following their work on the Web.

With that in mind, I wanted to take a moment to briefly revisit the two schools when it comes to image detection, watermarking and fingerprinting, discuss their pluses and look at which solution may be right for you.

Hopefully, by the end of this, you’ll be better prepared to decide on your own image protection scheme and know which technology to watch for.

The Two Schools

The challenge in finding copies of images on the Web has always been, to put it bluntly, that they aren’t text. Search engines, including image search engines, look for text as it is what is easiest to search for. Computers don’t understand what is inside an image (the reason CAPTCHAs are reasonably effective) and they have a hard time comparing one image to another. Scale that problem to the billions of images on the Web, and this issue becomes clear.

Solving the problem hasn’t been easy nor has it been perfect. However, there have been two different approaches that have been tried, both with great success but different ideal uses.

  1. Watermarking: Though most think of image watermarking as the placing of a visual mark over an image, usually a logo or name, Watermarking in this sense involves placing either an invisible or nearly-invisible watermark over an image that can only be detected by a machine. This method is used by Digimarc and (Note: The latter uses both for various purposes).
  2. Fingerprinting: Fingerprinting is an automated process that takes an existing image file, hashes it and converts it to a unique fingerprint. That string is, in turn, compared to other fingerprints for potential matches. This method is favored by PicScout and Tineye.

As you can see, the two systems are fundamentally different in terms of how they work and what they can do, these differences have an impact both in terms of how they are used and the situations they are best suited for.


The advantages of using a digital watermark are many-fold. First, they provide much more reliable detection. Where a very similar image by a different artist can trigger a fingerprint match, only copies of your actual image will trigger a watermark match. This also greatly helps with providing evidence for a potential court case since it can be shown exactly which image was copied from.

Watermarks can, typically, also survive great modification by users. Often images with watermarks can be detected even after cropping, rotation and distortion that would have caused a fingerprinting system to miss it (though fingerprints are getting much smarter about this as well). Some need only a few pixels to remain intact.

Also, additional information about the creator can be hidden in the watermark or stored in a database. This can help greatly with the worries surrounding the orphan works bills, which may allow certain uses of works if the author can not be reasonably identified.

In short, it works like image metadata in that it is invisible and is buried within the image, but it is much harder to remove and travels with the image even after it is modified and resaved.

However, watermarks can’t do anything to protect an image that’s already “in the wild”. Though you can re-release an image with a watermark, any copies of the non-watermarked image will not be detected. Also, watermarking systems tend to be expensive to use and, since one company usually controls the watermarks and the system for detecting them, they’re vulnerable to being rendered useless if the company closes.



The biggest asset of fingerprinting is that it works on any image, including one already on the Web. There’s no need to watermark the image or do any preparation work at all before publishing it, the technique works from the image itself.

Fingerprinting has become more and more accurate over the years, now competing with watermarks in that area, and is largely cheaper, including a free, effective offering from Tineye.

Also, since there are many fingerprinting systems in place and all work roughly the same way, its likely that you’ll be able to continue to match your images this way for many, many years to come. One company closing does not destroy your copyright protection system.

That being said, Fingerprinting does not, by itself, address the orphan works issue nor provide any proof of ownership, which is why PicScout recently released the Image IRC, nor is it always 100% reliable. Similar images, such as photos of similar subjects, have been known to trip fingerprinting matches, though the problem is nowhere near as severe was it was.

So Which to Use?

When deciding which to use, the easiest answer is to just say “both”. Since you can fingerprint a watermarked image, there’s nothing, save perhaps cost, that will prevent you from using both systems. Many already view watermarks as more of a copyright verification tool than an image detection one, making the two a natural combination.

However, duplicating efforts doesn’t make much sense. Due to refinements, both systems can do an extremely good job at finding matches, what is more important than the method is the individual system and how well it works. If it can find good matches and work as advertised, it’s probably a keeper, regardless of the system.

That being said though, you can’t even consider using a watermarking system if your images area already on the Web unprotected. It is only used, ideally, for new images as much of the usefulness is lost with unprotected copies being passed around. However, the added protections provided by watermarks make them worthwhile in many cases, especially when there is tight control over distribution and the images haven’t been leaked.

In short, if you have images that you are waiting to put on the Web or routinely put up new ones (and can justify the expense both in money and time), watermarking is likely a good solution due to its added benefits. For those who have their works online already or don’t wish to invest the time/money required, fingerprint matching, via Tineye, is free.

However, even that distinction is being blurred by PicScout’s Image IRC, which is a database of fingerprinted images, which provides something of a “best of both worlds” solution. The problem though is that it, right now, is targeted solely at large stock agencies. We’ll have to see later how approachable it is by smaller artists once it is finally released.

Bottom Line

In the end, any robust image protection system will likely use at least some of both. If your images are valuable, trusting any one solution by itself to handle detection and orphan works prevention is risky. Multiple tools are, most likely, the way to go when practical.

But the most important thing, as mentioned above, is not that the system works, not the technology behind it. If it works well, it doesn’t matter if it uses fingerprints or watermarks. That is where the real challenge lies for these companies.

So, with that in mind, I’m looking forward to, over the next few months, talking more about developments in this field and how they might impact visual artists of all kinds and all backgrounds.Image Detection: Watermarking vs. Fingerprinting Image

Want to Republish this Article? Request Permission Here. It's Free.

Have a Plagiarism Problem?

Need an expert witness, plagiarism analyst or content enforcer?
Check out our Consulting Website