Google automatically remove watermark from stock photos

Google shows how easy it is for software to remove watermarks from photos

image from google

Photography professionals will often slap watermarks on their images to protect their copyrights and prevent people from using them without their permission. However, the researchers were able to identify a glaring error in this approach and exploit it to negate the visibility of watermarks altogether. The method – which has been documented in more detail in a paper titled On the Effectiveness of Visible Watermarks [PDF] – essentially takes advantage of the fact that watermarks, by design, are applied to photos in the exact same manner. While this makes it easier for stock photography markets to establish their brand, Googlers were able to leverage this consistency in order to develop a computer algorithm that is capable of autonomously editing out signature marks out of images.

It’s much easier to remove a watermark from a photo if you can extract a copy of those overlaid graphics first. So what Google’s software does is first scan thousands of sample images from a given stock photo provider, comparing each one until it detects a repeating pattern that inevitably reveals the watermark being used. Knowing what the watermark looks like then allows the software to remove it, leaving a clean photo that’s almost impossible to distinguish from the original copy.

The drawback to this approach is that it really only works on photos sourced from a stock photography provider with a massive catalog of watermarked samples that can be analyzed, but more often than not it’s those services that people are trying to cheat. Google didn’t publish these findings as a way to help the internet steal from talented photographers, however. The company instead took this research one step further and found that by slightly warping a standard watermark from image to image, even the smallest of deformations made it impossible to automatically remove the watermark without leaving obvious artifacts behind. So by adopting this modified approach, stock photography providers, and even professional photographers, can ensure that their imagery can’t easily be used without their permission.

The team behind the watermark-removal algorithm was able to train software with enough public examples to identify watermark patterns and then, through a process called “multi-image matting,” separate the watermark’s components from the rest of the image. Then, because the software understands the elements of the watermark like its opacity, structure, and shadow or color gradient effects, Google’s algorithm is able to remove it from any photo containing that specific watermark or a similar one.

To fix this, and create stronger copyright protections for images on the web, the team suggests adding elements of specific randomness to the watermark. However, you can’t simply change the location, or make changes to the opacity of the watermark, Dekel and Rubinstein explain. Instead, you need to make changes that will leave visible artifacts after the removal process. This includes adding “random geometric perturbations to the watermark” — effectively warping the text and logos being used. That way, when algorithms like the one Google uses try to scrub the watermark out, they’ll leave outlines of the image because these systems are trained to look for consistency and work by targeting the vulnerabilities inherent in that consistency. The team admits that the defense isn’t a perfect one. There will likely always be more sophisticated algorithms developed to bypass current practices, in a cat-and-mouse struggle similar to that of cybersecurity protections. However, the current state of watermarks leaves image protection in a sad state, they say, and even just a little bit of the right kind of randomness can go a long way in keeping photographs safe from theft in the short term.