Is enhancement the same as manipulation?
How far can you enhance an image or video before you cross the line into manipulation?
The UK is currently prosecuting two men accused of a crime. Part of the prosecution's evidence is a video0. In showing it to the jury, the prosecution have said:
the two minute and 41 second-long video is "extremely dark" but the "unmistakeable" noise of a chainsaw can be heard followed by the sound of a tree falling.
Police experts have "enhanced" the video as much as possible but it has "not been interfered with", Mr Wright tells the jury.
I think most reasonable people would agree that creating an AI "Deep Fake" by inserting the faces of the pair into the video, would be unacceptable.
What about boosting the brightness on the video? That seems pretty unobjectionable to me and, I suspect, most neutral parties.
Suppose the prosecutors used AI to enhance the image? Perhaps adding a background which wasn't there up maybe upscaling the video resolution and introducing elements which didn't exist before? I think that's a step too far. Algorithmic enhancement strays into manipulation territory.
But what if the police ran a face detection algorithm on the video and only boosted the visibility of those parts, rather than the rest of the video? Now I think we're haggling over price.
The photographer Paul Clarke has a wonderful blog post about enhancing photographs of MPs - take a look at those photos. Are they enhanced or manipulated? Do you feel differently if it is a photo of an MP from "your" side?
But just brightening and colour correcting is fine, right?
This is a well-known problem in legal circles1. Boosting the colouring of a photo may make an injury seem more severe. Zooming or cropping an image may make someone seem closer to the action than they were.
The Crown Prosecution Service has this to say about video2 evidence:
In terms of proving the authenticity of the video recording, the Prosecution must be able to show that the video film produced in evidence is the original video recording or an authentic copy of the original and show that it has not been tampered with.
I suppose it's pretty easy to show that the produced evidence can be derived by taking the original and twisting the brightness and contrast knobs. I also guess that the defence could bring in an image manipulation specialist to show that the enhanced version introduces unacceptable changes.
Although that brings with it some problems about whether an expert in manipulation can say they're an expert about the contents of the media. (No, basically.)
I'll leave you with these words from a House of Lords report in 1998:
The existence of a technology that can be used to modify images in this way need in itself be of no great concern; even the widespread availability of the technology at low cost might not cause concern.
But an apparent lack of understanding of the implications of both these facts should cause concern and warrants further study. The public and all those in the legal profession should be made more aware of the technology, what it can do, and what its limitations are.
It was suggested that criminal convictions that were dependent on evidence captured by digital cameras could be at risk if defence lawyers began to realise how vulnerable such images are to manipulation.
The trial continues.
Update!
The initial video was totally dark, with just the sound of wind and a chainsaw leading up to a giant crash.
A second version has now been shown to the jury, which has been enhanced by a Northumbria Police digital media examiner.
The contrast has been changed, a white border has been put around it and the image has been made brighter.
Here's a clip of the enhanced version:
If you were presented evidence of a completely dark video, how could you be sure that subsequent "brighter" version was derived from the original?
-
To be clear, I'm not at the trial. ↩︎
-
With thanks to several anonymous legal friends for pointing me in the right direction. ↩︎
-
There's a good discussion about the admissibility of video evidence in [2002] EWCA Crim 2373 ↩︎
The problem is only due to get worse with manipulated videos/images: with both Google and Apple having "AI Intelligence" in their cameras (to the extent of Samsung phones automatically "faking" pictures taken of the moon), "you" can present the image as the one you took, totally original and that it had not been tampered with - but the manipulation happened "in the background" without your knowledge the instant the photo/video was taken...Should the photo/video still be submissible?
@Edent
If I got the video in binary form, I could read enough of the #ffmpeg man page to adjust the brightness and contrast myself. The results wouldn't be as good as a specialist could produce, certainly, but they'd be enough to tell me whether the enhanced video could plausibly have been derived from the original.
For non-geeks, my answer is quite different: we can't be sure of any evidence presented in court, because witnesses, police officers and experts could be lying. That's why we have perjury laws with such stiff penalties. If you can't get your hands on the evidence yourself, you ultimately have to trust the system.
ffmpeg
| Reply to original comment on infosec.space
@Edent
Very interesting!
i would imagine, as you have said, that changing the brightness and contrast should be admissible, as the human eye can usually see a lot more in the dark then a standard camera lens. Colour changes though potentially evoke different reactions in folk. Tricky!
| Reply to original comment on seocommunity.social
James A says:
One way to reduce risk and controversy could be to provide the original video and each of the software processing steps that were involved to enhance it. That could allow people to gain confidence in the final video by independently inspecting, critiquing, and, if necessary, replicating, each of the steps involved to create it.
More comments on Mastodon.