This is the first test of the webcam, when loosely placed within the Photonic Instrument's integration sphere. The sphere was dimly lit with a small single AAA battery LED hand torch. There is still no coating on the inside of the sphere, but the illumination appears relatively uniform. At the moment, the webcam is intact bar the IR illuminator having been disabled. There are plans to remove the camera lens in the future to get an even more uniform light distribution on the CMOS sensor.
This is the unaltered 640 x 480 VGA output JPEG from the camera ... And no, it's not just a black square - look closely. You'll be staring into the face of Randomness. All hail Randomness. Einstein was wrong about that game of chance. There is a long list of possible sources that contribute noise to a webcam image, but our Photonic Instrument will rely primarily on photon shot noise and Nyquist thermal noise. If you stare for a few hours at this image, you will see slight colour variations. This is despite the illumination source being pure white. The individual colour shifts occur as a result of the signal to noise ratio (SNR) being so low that Gaussian and Poisson stochastic fluctuations at individual photo detector sites are significant. The significance is such that it manifests itself visually as perceptible variations in colour.
The entire principle that underpins the operation of the Photonic Instrument is that this image is truly random and unique, only influenced by the Four Elements and God. Not wishing to overly dramatise, but it is totally unpredictable by human kind. The NSA, FSB, Mossad and the BBC Weather Service cannot predict this pixel pattern exactly. Since the images will never be displayed anywhere on a monitor during production, even TEMPEST surveillance might prove ineffective. It really is God playing with his dice.
At Headquarters, we were all really really excited when we saw this. Dom purred out 'Spiffing, what?' Helen raised an eyebrow, and Paul drank lots of laboratory cleaner till he passed out.
A great deal of image analysis hasn't been performed upon this JPEG, but some is shown below. The coloured squares show the pixel distribution across a small part of the JPEG. This section has been equalised to emphasise the per pixel variations. The raw JPEG file (1 webcam frame) is approximately 20 KB in size, and this sets the expectation for what we will have to work with for randomness extraction. Our sense is that we might be able to extract a quarter of the raw JPEG files as full entropy, that being about 5 KB per JPEG frame. We are researching what we believe to be a unique randomness extraction algorithm.
The variations appear totally random (within the context of the JPEG algorithm that produced the image). The 8 x 8 pixel sub images are a typical characteristic of all JPEG images.
The histogram of the raw JPEG shows pixel values ranging from 0 to 30 in the standard normal distribution, albeit with a substantial Kurtosis. The other observable point is the mean. A mean of 10.8 is approximately 1 1/2 stops below mid grey. This is confirmation that the webcam's auto exposure function has been unable to compensate for the very low illumination. This failing works to our advantage in minimising the SNR, and therefore maximising sensor noise.
Counting the unique colours in the JPEG also gives the first true indication of the severity of the one true enemy of all ip camera based chaotic random number generator maker people. The dreaded JPEG algorithm. All of the image variation is due to in camera random processes as the integration sphere is featureless and plain white. Thus a VGA JPEG image with such a pixel value distribution might be expected to contain in the region of 28,830 unique colours (20 * 31 *31). The file only contains 2060 colours. 93% of the theoretical colour entropy has been optimised out of the image. That's good compression!
It is our intention that randomness extraction will take place from the raw JPEG files, without opening them. It is a common fallacy to believe that image files have to be opened for randomness extraction to take place. We will demonstrate shortly that decoding and opening up a JPEG image just creates a great deal of pseudo random entropy that most believe is random. It is not, and is a direct consequence of the output from the JPEG decoding algorithm. The true entropy is the file itself. This is a perfect example of the age old philosophical question we've all pondered, "How much of sample variability is entropy, how much is just complexity?". The file content is only shown here to satisfy ourselves that the Photonic Instrument can be made to work, and to illustrate a facsimile of the captured sensor noise.
In the spirit of the Blitz and internet outages, we will carry on ...
[#1] John Kelsey, Entropy and Entropy Sources in X9.82, NIST, July 2004
Add new attachment
List of attachments
|Kind||Attachment Name||Size||Version||Date Modified||Author||Change note|
|colour-of-light.png||7.9 kB||1||25-Sep-2015 00:34||cossoft|
|histogram-spread.png||2.3 kB||1||25-Sep-2015 02:07||cossoft|
|photonic-test-histogram.png||19.3 kB||1||25-Sep-2015 01:15||cossoft|
|photonic-test.jpg||20.4 kB||1||25-Sep-2015 00:33||cossoft|