When there is sensitive information that you want to keep private in a photo, such as a credit card number or a face, users can blur or scribble on it. However, that is no longer a secure way to protect information.
According to Howtogeek, people often blur sensitive information in screenshots or regular photos for a long time. But as technology advances, it becomes much easier to crack blurry images
Researchers at the University of Texas and Cornell University (USA) recently put a simple deep-learning algorithm to the task of identifying blurred faces. As a result, the algorithm provides an accuracy of 71 – 83% when making 5 guesses. This is very alarming.
In fact, tools like this have been around for a while and aren’t too difficult to use. Many of the same tools have been trained to recognize blurred faces using YouTube’s built-in video tools, with all the user needs is some open-source software and a bunch of blurred photos to train It
It’s not hard to see how this same technology can be used for more sensitive information, such as credit card numbers, license plates, social security numbers, etc., but fooling machine learning algorithms is another story.
As a result, blurring text and photos is no longer the best way to remove sensitive information, but there’s so much more to it. Sometimes, scribbling things also raises security concerns. In the past, the CEO of Sony’s PlayStation team has leaked secret documents with a series of numbers, dates, and figures covered with a black Sharpie pen. But when those documents are scanned, it’s pretty easy to see the text below the marker. Now people can see details of Sony’s business.
For the same reason, users should avoid using Sharpie both in the physical and digital worlds. Blurring or drawing lines that obscure information will not completely block information. In the case of digital tools, sometimes users only create a layer on top of the image they want to cover, but this layer can be removed if handled carelessly.
In order to secure sensitive information, users are advised to ensure that no traces are left on images in more advanced ways. Example with the Apple Card image in the article. Name blurring may not be readable to the naked eye but can be handled by a machine learning engine. So the best way is to block the name completely with a solid color bar.
