Today’s image processing technology can not only restore a highly pixelated image but even extract text from it.
This week, Dan Petro, a lead researcher at security firm Bishop Fox, said he had clearly recovered the letter information from pixelated text images.
Last year, Jumpsec Labs shared an open challenge, hoping someone would be able to recover the textual information in the image below.
Petrol came up with a solution after researching various pixelation and de-obfuscation techniques, which resulted in a clear reproduction of the text, but at Jumpsec’s request, he only disclosed part of it.
While some techniques exist for enhancing pixelated photos of people or landscapes, such as Google Brain, there is no mature solution that can truly restore text in pixelated images with complete accuracy, while removing unwanted impurities.
Google Brain restores the image (middle) from the provided pixelated image (left), the right column shows the actual image before pixelation.
Petrol’s success in cracking the Jumpsec challenge prompted him to release a new open-source tool called Unredacter on GitHub with Bishop Fox and to publish his detailed findings in a Bishop Fox blog post.
Although Petrol said that these technologies are not as good as imagined in actual application scenarios, and there are still many shortcomings, it also sounded the alarm for the behavior of processing sensitive information in a pixelated way. He believes that using methods such as black bars to directly cover sensitive information is much safer than pixelating or blurring.
In addition, the researchers also believe that when editing text, it should be edited as an image, rather than masked with simple HTML/CSS styles. For example, text masked with the same background color as the body of the text itself can be easily shown when highlighted.
