@Lesnums) Posted on 23/08/17 at 09:30 Share:
The giant Google has almost unlimited research and development resources and dedicates entire teams to image processing. Tali Dekel and Michale Rubinstein have just proven that it is entirely possible to automatically disappear, thanks to an algorithm, a large majority of the famous "watermarks" or digital watermarks meant to protect online images from misuse.
The purpose of this research is not, according to Google, to want to naturally blow up all the watermarks that could appear in Google Image and to improve the visual experience of Internet users. This is to warn the authors and structures that use them of the vulnerability of this technology and give them ways to make them more effective. Finally, it is clear that Google is working on the subject and that, if one day the desire took it to use such an algorithm on a large scale online, it could do so...
Why is it finally quite simple to remove it?
The very design of a visible watermark is to deposit (and crush) on the image an external visual composed of more or less large and more or less complex elements to make impossible (degraded in reality) the use on which it is applied. Manually removing a visible watermark can be time-consuming, tedious, and require advanced image editing skills. Automatically detecting and correcting an image is thus very difficult. It is therefore, on paper, a good security. The problem is that the same watermark is used in the same way on a considerable number of images. Take for example those used by image banks such as Adobe Stock, Getty or Shutterstock. In this case, it is possible, thanks to an algorithm and automatically, to take advantage of this profusion of images tattooed in the same way and to reverse the effect in a way. The algorithm, by dint of always studying the same watermark deposited on very different images, can "learn" all the possible scenarios and thus remove it very easily thanks in particular to a mass study on the opacity.
How to strengthen the security of your watermarks?
To make a watermark less vulnerable, all you have to do is modify it as regularly as possible. Changing it in size and position is not enough. It is necessary, to a certain extent, to introduce random geometric shapes into each watermark. This is equivalent to making a single tattoo per image.
Be careful, this solution does not constitute an absolute guarantee to protect your images, but it will make the life of automatic deletion algorithms much more difficult.
