Image result for pixelated person

It is a common practice to pixelate images that are uploaded to the internet to protect their content or simply hide a specific object that appears in the image. However, in this new era in which machine learning is becoming so powerful, it seems that pixelating an image is not a reliable method anymore if you want to preserve the identity of a person or information showing onscreen.

It turns out that researchers at the University of Texas and Cornell have created an algorithm that successfully identifies faces, objects, and even letters and numbers that have been pixelated. Advances in machine learning methods are becoming so effective that simple security protocols are becoming obsolete in the face of advanced AI. What is most surprising about this is that engineers only have to apply some principles of these technologies that already exist to positively identify the contents of a pixelated image.

So, how accurate is it?

In the images used for their tests, the system devised by these researchers was able to identify a pixelated face with an accuracy of 70% (probability rises to 83% in five attempts) compared to a probability of 0.19% by a human in the same test. The images for these tests have been taken from YouTube, which are seen by thousands, even millions of persons around the world.

It is important to notice that this technology does not turn a pixelated image into a sharp image as we see in movies, it simply identifies the content that is pixelated.

The purpose of this research is to raise awareness about the risk of just pixelating an image to hide something. Media and news companies use this simple technique regularly to protect the privacy of people shown on camera or to hide sensitive information or graphic content, and video platforms such as YouTube even offer a blurring feature that tracks a face and automatically blurs it.

What can you do about it?

With all this being said, what can you do to completely protect and hide something in a video or image? There may be more sophisticated encryption techniques to completely hide something in a picture, but the easiest way to do it and that anybody can do it is simply by putting a black box over the person or object you want to hide, this way there is no information available in that area for a hacker to obtain. It may be less pleasing to see a black box instead of a simple, non-distracting blur in a picture, but it is the easiest and most effective way to avoid exposing sensitive information that you don´t want to show in your image or video.

Source :


Kevin Nether