Last week Google released a research that involved removing watermarks from photographs posted on the Internet, according to the company they can be removed easily through programs that use artificial intelligence algorithms.
The company revealed a document during the 2017 Conference on Computer Vision and Pattern Recognition called "On The Effectiveness Of Visible Watermarks" which demonstrated that the company worked with the logic of separating the watermark from the photo as if they were two layers. . Subsequently, the mark is recognized as a dirt and is sanded until it is completely gone.
The developed algorithm works with a simple logic. The company's main targets were large image banks found on the internet, such as Adobe Stock, CarStock and Fotolia.
The Web Giant motivated the Shutterstock company to make some changes, from now on, the company will adopt non-standard watermarks, which use deep learning to ensure protection security. What changes with that? Non-standard images make it difficult for these algorithms used by Google to work.
According to Shutterstock's CTO, Martin Brodbeck, in an interview with The Next Web, "The challenge was to protect the images without degrading the image quality. Changing the opacity and location of a brand does not make it any safer, however, modifying its geometry, yes."
See in the image below how the company's watermarks were with the changes.
Google recommendations
The Shutterstock company's was taken as a precaution after recommendations from the Search Giant. Google researchers concluded that using the same watermark pattern made it easier to identify by algorithms that always looked for the mark in a precise, standard way. The removal then started to be done almost automatically, in just a few moments.
With that, the experts suggested minor changes to the brand format to make it difficult for the algorithms to process, which would have to identify a larger area to account for all the variations. Therefore, the removal of the mark would not be total.
