Thread by @spdustin

With the bounty of raw video and photo files available from the January 6th Capitol terror attack, participants can potentially be identified using the video taken by someone else. Or, more specifically, by identifying who is recording the video.

A thread!

1/9

Every cellphone uses a sensor to convert a visible image (or video) into digital data that is processed to create the final media file. These sensors—and the digital signal they create—are imperfect (because physics). But every sensor is *uniquely imperfect*.

2/9

These imperfections result in digital “noise” in the data. Videos taken by one cellphone can be analyzed to extract a sort of “fingerprint” of the digital noise inherent in that cellphone’s imperfect image sensor. That fingerprint persists, even in videos taken weeks apart.

3/9

If an analyst has physical access to the cellphone thought to have recorded a video, a new video can be taken, and its noise fingerprint compared to the video under investigation. If the fingerprints match, you’ve identified the source camera.

It gets more interesting…

4/9

There are even analysis techniques that can extract a “partial” noise fingerprint from a still image, and identify a match between a still photograph and a video—even if the original media was uploaded to a social media and compressed. Why is this so relevant?

5/9

Let’s say “John Q. MAGA” live streamed his exploits to DLive. He’s wearing a mask, so you can’t identify him.

Let’s also say that “John Smith” is on YouTube, where he used the same cell to record an (unmasked) video of himself.

Reviewing testosterone supplements, probably.

6/9

Now, you don’t know if “John Q. MAGA” is actually “John Smith” by sight—he’s wearing a balaclava, after all. But multimedia forensics can extract the sensor noise fingerprint of the two videos, and if it was the same phone, those fingerprints will match. Boom: unmasked MAGA.

7/9

There are many papers detailing various approaches that multimedia forensics experts use to link source videos across social media platforms using the sensor noise fingerprint. Here’s one:

8/9

ncbi.nlm.nih.gov/pmc/articles/P…

The money quote from that paper: “its effectiveness has been proved to link Facebook images to YouTube videos…”

I just wanted folks to be aware that, just because a cellphone camera operator hides their face in their criminal selfie videos, there are ways to identify them.

9/9

PS: to tie it back to the first tweet in that thread…

If you can use sensor noise to identify a person recording *someone else committing a crime*, you’ve just opened up a new avenue of investigation; one that can place the videographer alongside the subject of the video.

P.P.S.: remember that Parler gave verified status to anyone who sent a photo of their drivers license?

How do you think they took that photo?

Using their cellphone camera.

Which adds that unique sensor noise fingerprint.

So, yeah…there’s a lot of data available to the FBI.