Video Fakes, Deepfakes and Forgeries

---

You’ve seen those videos on social media with a friend’s face superimposed over an actor in a famous movie, right?  If not, take a look at this video.

Advancements in video technologies such as artificial intelligence are making these funny, but worrying, deepfakes easily and quickly achievable by anyone with access to a smartphone or computer.   It doesn’t take a huge stretch of the imagination to think of ways this could be used maliciously.  Luckily, with expert knowledge in forensic video analysis, there are techniques to help root out the truth behind videos suspected of forgery.

Digital camera sensors have a unique hidden sensor noise pattern which isn’t noticed without filtering out various values of the image. Think of it like a thumbprint; It is not noticeable until you look very closely, and it is unique to each individual. Over time, the sensor noise pattern should be consistent throughout the duration of the video.

Any inconsistencies in the pattern, such as a computer-generated overlay of someone else’s face, would disrupt this sensor noise thumbprint inherently present in the original video clip, therefore making a convincing argument of tampering.  Other techniques involve looking at changes in compression artifacts as well as inconsistent changes in motion or brightness.  That being said, those issues with motion or brightness inconsistencies are not always present.