Hacker News Reader

Comment by yaur | Hacker News Reader

> So, even if you used two cameras of the same make and model to snap a uniformly lit surface – where every point on the surface has the same brightness – there'd be subtle differences unique to each camera.

Ok, but will that survive JPEG compression? Since JPEG is meant to remove details that aren't visible to the human eye it would be surprising if this fingerprint survived in a usable way.

View on HN

Yes, it does survive compression. That’s why if you’re into astrophotography and want to stack pictures you start by taking a black picture. That way, you capture the noise pattern to remove from the noise from the other pictures.

Getting rid of dark noise is one thing, the gain of each pixel is apparently a bit random due to manufacturing processes as well. You'd have to take a picture of a uniform white surface to calibrate that out as well.

Yes, exactly. Except that we take a picture with no light at all to get an image of the noise from the sensor, typically leaving the lens cap on.

Wouldn't astrophotography be done using lossless (or no) compression?

It depends. Cheap webcams are commonly used for planets, for example. They record a bunch of JPEGs or a movie, which are then processed and stacked. The idea is that with many exposures you get as much signal as with a long one, without the associated artefacts (noise from high sensitivity, blur from fluctuations in the atmosphere, etc). Compression artefacts don’t matter as much, but sensor noise does.

Of course, in some cases shooting RAW from a specialised camera is better, but quite often JPEGs straight out of a standard (cheap DSLR or mirrorless, or webcam) camera are more than enough. Takes an order of magnitude less memory as well, which counts when you have stacks with 200 pictures.

There is a discussion here, for example: https://www.planetguide.net/astrophotography-webcam/

The effect is real, but cameras actively try to suppress it. Cameras have a map of particularly bad pixels which are ignored (much like a bad sector map in a hard drive), and when doing long exposures most cameras will/can do two exposures, where the second one is made with the shutter closed.

I'm skeptical this works after boiling 24...50 megapixels down into a tiny web-sized image with 1-3 MP.

According to the article:

> Technically, it should be possible to suppress, for example, by reducing the image resolution, says Farid. But, by how much? This of course depends on many factors such as the type of device used for image capture, as well as the fingerprint matching algorithm employed. There is no one-size-fits-all solution to fingerprint removal.

JPEG compression is calculated locally in small segments, so variations across the whole surface can still survive, each segment being independent. JPEG is not "meant" to remove detail, more like prioritize large-scale phenomena, which in this case would preserve imbalances between segments.

Reply on news.ycombinator.com