SAN FRANCISCO (KCBS) — Pixelation has been used for quite a while as a way for television to conceal someone’s identity, but researchers have found a way to get around it – and it’s not that difficult to do.
Humans aren’t the only recognition masters anymore. As computer vision becomes more sophisticated, it’s seeing more things. But it’s not like in the movies where you press a button and uncover the image either.
Researchers started by creating a machine learning model. Training a computer with a set of example data rather than programming it. They then presented it with an image that’s been pixelated or blurred.
“The obfuscated image can give you a label about the image. They can say, ‘this is a cat,’ or, ‘this is a picture of George Clooney,’” Richard McPherson, a PhD student at the University of Texas studying privacy and security said.
Richardson is also the primary author of the study. He says that with their method, there have to be some suspected elements that you believe are in the picture.
“Our technique doesn’t allow you to take a picture of someone you have no idea about, and recover who that person is. You have to have some basic assumption of where to start,” Richardson said.
For those worried about bad guys adopting this relatively simple technique – the more extensive the obfuscation, the lower the rate of success.