[Cryptography] e: The ultimate random source

Arnold Reinhold agr at me.com
Thu Feb 20 22:58:16 EST 2014

On Wed, 19 Feb 2014 21:50 Peter Gutmann wrote:

> Christian Huitema <huitema at huitema.net> writes:
>> If you rely on a capped camera to generate white noise, you may be out for a
>> surprise with at least some cameras. There is a lot of filtering and
>> processing that happens on board the camera itself, e.g. conversion from
>> Bayer pattern to YUV or RGB, firmware that enhances the image, compression to
>> JPEG before transmission on the USB bus, cropping and resizing on demand, etc.
>> I would not be surprised if some cameras, when capped, just transmit a black
>> image.
> I would certainly hope they transmit as close to a black image as possible,
> since they're designed to do that.  For example when shooting dark scenes some
> (still-image) cameras will take a second image with the shutter closed and use
> that to get an idea of any noise present, then use that information to cancel
> noise in the primary image.  This will remove a lot of the hoped-for noise. In
> addition since a lot of what you're hoping for is thermal noise, taking the
> images infrequently (so the sensor is relatively cool, therefore with low
> noise), or with the camera in a cold room, is going to reduce noise
> significantly.  This is why devices like telescopes and medical imaging
> devices use externally-cooled CCD sensors, I've seen serious (or perhaps
> slightly crazy) people pull apart thousand-dollar DSLRs to fit TEC (Peltier)
> elements to them.

Umm, it's more complicated than that. Here is a resource that explains some of what is going on: 


The primary purpose of cooling is to reduce dark current, not noise.  Subtracting a dark image removes the dark current bias, but it *increases* noise.  Thermal noise is reduced by lower temperature, but classic thermal noise power is 4kTdeltaF, where T is absolute temperature, = Celcius + 273. So going from 0C to 80C is only a 30% difference.

Note that there are other sources of noise, including shot noise and noise associated with any amplification. Shot noise is higher for higher light levels, another reason id question the lens-cap-on approach.

> I've already said this before, but I'll say it here again: The most you can
> say about sensor noise (optical, audio) is that any value you get for your
> particular setup applies only for that specific combination of hardware,
> software, and environmental conditions.  These devices are specifically
> designed to get rid of as much noise as is physically possible, so can't be
> relied on, in isolation, as noise sources.

The question is, can one presume at least 256 bits of noise in a megapixel camera image of a complex scene for any camera with at least, say 10 bits of gray scale. Removing "as much noise as is physically possible" is the key phrase here. I'd bet yes, but it deserves more careful study with input for camera and image processing experts. 

Arnold Reinhold
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.metzdowd.com/pipermail/cryptography/attachments/20140220/23a2272e/attachment.html>

More information about the cryptography mailing list