Artificial Intelligence Can Soon Make Pixelated Image Reversal Possible

Do you remember the old days when if anything was to be censored for privacy reasons, editors used to put up a colored box over it? To date, people also find it as a good practice to hide details in screenshots e.g blacking out the name, picture or personal information of anyone to protect them from being targeted by fanatics.

However, over the years another effect called pixelation has also emerged to achieve the same privacy goal. The good thing about pixelation or blurred effect is that it reveals a major side of what’s going on while also taking care of protecting the person’s identity. This technique has been more popular to hide body parts for reasons.

There are also high chances that you may have seen a practical implication of pixelation during the massive protests that have been going around the world lately. Law enforcement agencies use the footage to identify participants so that they could either take legal action against them or put them in the record for participating in a violent protest.

Looking at its usage and benefits, researchers are now working on developing the sort of AI technology that can actually reverse the pixelation effect to make users easily identify the faces hidden behind those blurry image.

A glimpse on such a technology has been given by a group of scientists from Duke University as they published a paper based on Photo Upsampling via Latent Space Exploration (PULSE) system.

According to the researchers, the main aim of a single-image super-resolution is to build a high-resolution image from the corresponding low-resolution (LR) input.
This AI Can Turn Pixelated Photos Into Recognizable Faces
The end result is pretty much similar to how you see in movies and TV shows when a person from investigation authorities orders to “enhance” the image quality and you get a clear footage of the person hidden behind the pixels. Well, yes what you used to see in science fiction, PULSE can pretty much turn it into reality.

The team of researchers also did demo tests in which they downscaled the images of selected people and passed them through PULSE. The final look of the image coming from PULSE was not exactly like the original image of the person but still, the results were very close.

The only problem with facial recognition AI however still remains and that is how false positives are more likely to appear than accurate results and that is something PULSE will have to continue working on.

Read next: Facebook reveals the results of the Deepfake Detection Challenge and shows some promising prospects
Previous Post Next Post