What Enlarging a Video Frame Adds: The Rittenhouse Drone Footage

On 10 November 2021, in a Kenosha County courtroom in Wisconsin, the prosecution asked to do something we all do ten times a day: pinch an iPad screen to make a picture bigger. The defense objected. The judge did not rule on the science. He demanded proof, and the prosecution could not produce it.

“The video speaks for itself.” Litigators say it often. In Kenosha it turned out that the video says only what its encoding, its copy and its viewing tool allow it to say.

Two people were killed and one wounded during the Kenosha unrest of August 2020. This piece is not about guilt. It is about a file.

What actually happened

The case is State v. Kyle Rittenhouse, in Kenosha County Circuit Court, Judge Bruce Schroeder presiding. The first fight over the drone footage came on 10 November 2021.

Prosecutors wanted to enlarge the footage for the jury with pinch-to-zoom. Defense attorney Mark Richards argued that iPads contain “artificial intelligence” that renders things through three dimensions and logarithms, and that the jury would see Apple’s programming’s guess rather than what was actually there.

Schroeder sustained the objection without ruling on the science of interpolation. His ruling was narrower: to zoom, prosecutors first had to put on an expert testifying that enlargement would not alter the image. During a short recess they could not produce one, and showed the jury the footage as originally recorded, with no zoom.

The defense’s phrasing was wrong. Pinch-to-zoom scales an image. It does not fabricate content or run “artificial intelligence” on it, and technology journalists said so that same week. Still, there was a real point under the bad phrasing. Forensic video expert Grant Fredericks told the AP that pinch-and-zoom cannot be replicated exactly, and that the defense had raised a legitimate concern. Computer scientist David Forsyth told the same reporter there is “no way” enlargement creates objects or people, but that every method adds pixels algorithmically.

So a loosely argued objection won, because the prosecution had no witness to explain what the tool does and no record showing the enlargement could be reproduced.

A second dispute concerned the file itself. Prosecutors received the drone video on Friday 5 November 2021 and sent the defense a copy the same day. That copy was compressed, at lower quality. Natalie Wisco, a defense assistant, said the full-quality file came only after she gave prosecutors a flash drive (discovery, resolved by sneakernet).

According to reporting on the defense’s own motion, its copy was about 3.6 MB at 480 x 212 pixels against the state’s 11.2 MB at 1920 x 844. Those are the defense’s figures, not established fact.

Assistant District Attorney James Kraus said the file was too large to email and that the compression happened in transfer, which he explained as an iPhone-to-Android transfer: “We didn’t compress anything, we didn’t change anything.” He added: “We’re focusing too heavily on a technological glitch.”

On 17 November 2021, the day after the jury began deliberating, the defense moved for a mistrial over the video quality and disclosure. Schroeder did not rule immediately. He said he had qualms about the video, and he let jurors view it on a computer in the jury room: “if they’ve got everything correct and reliable, then they won’t have a problem. If it isn’t, it’s going to be ugly.”

On 19 November 2021 Rittenhouse was acquitted on all five counts. The motion, on which Schroeder had not ruled, became moot, exactly as the AP had anticipated. The hardest question about the file was never answered.

The artifact: a video file that has been through several lossy steps

The file is not “what the drone saw.” It is what remains after a chain of engineering decisions. What follows is how consumer and drone video is typically stored, not a claim about the Kenosha file.

Compression. Video like this is usually stored with lossy compression, for example H.264 or its successor HEVC. The codec splits each frame into blocks, applies a DCT-style transform, and quantises the coefficients so fine detail is discarded. Colour is stored at lower resolution than brightness, typically 4:2:0. Most frames are not full pictures at all. They are motion-compensated differences from earlier or later frames (P and B frames), inside a group of pictures that opens on a full I-frame.

When the bitrate is starved, small, fast-moving objects suffer most: blocking, ringing around edges, smearing along the direction of motion, and ghosting. A person seen from a drone may be only a few macroblocks across.

Enlargement. A resampling algorithm such as nearest-neighbour, bilinear, bicubic or Lanczos computes values for pixels the sensor never measured. The first copies the nearest pixel. The others take a weighted average of the neighbourhood. None of them adds information. All of them add plausible-looking guesses. The result is smooth, but smooth is not recovered detail.

Pinch-to-zoom does not run “AI,” but any enlargement is resampling, and resampling invents intermediate pixel values. So “the video speaks for itself” is false.

What an expert witness checks when the evidence is video

An expert does not argue about what the picture looks like. They build the record that lets a court decide.

First, where each copy came from: provenance and a hash for every one. Then a comparison across the disclosed copies of resolution, bitrate, codec, frame rate and container metadata. Then the enhancement pipeline, documented so it can be reproduced: the tool, the interpolation algorithm, the scale factor, any sharpening, and the order of the steps. Last, in writing, exactly what can and cannot be concluded from an enlarged frame.

That is what prosecutors in Kenosha could not produce during the short recess.

Why it matters beyond this case

Israeli criminal and civil cases increasingly turn on video: shop cameras, dashcams, bystander phones. I know of no Israeli ruling on this point, and I will not guess at one. But both questions from the American case transfer directly. Do all sides hold the same copy at the same quality? And what does “enlarged” actually mean? Someone should be able to answer before the judge asks. That is the work of a source code expert witness.

The above is general information and not legal advice. Case facts are drawn from the sources listed.