LSB + von Neumann passed 
And I think LSB + von Neumann is a good approach for preserving the purity of quantum randomness. It’s very slow though. But you can use as many pixels as you can. However, I recommend skipping at least every 4 pixels to prevent cross-correlation caused by the demosaicing algorithm that interpolates the color value of a pixel based on neighboring pixels.
The “sample” or “word” or “byte” in my method (the third method) is basically the current produced by the photodiode. Higher value = higher number of photons detected, lower value = lower number of photons detected
I note you only used about 2 seconds of data (6.82 Mb tested with 3.85 Mb throughput). This is a very short test sequence, and not even long enough to complete a single OQSO test.
I suggest a minimum of 100 times that (a few minutes), but more data tested is usually better. Some other testers suggest 10MB (80Mb) as a minimum sample.
Yeah, my webcam-based noise source sampling rate maxed out at ~7.5 Hz on full resolution and I’m only using a small number of pixels. I can keep the script running so I can test the generated file using tests that need more data.
Btw the QNG Meter shows only 2 seconds because I tested a pre-generated file. To generate that file actually takes a few hours.
I thought that might be the case. A 24-hour sample would give a much clearer picture of the statistical quality.
When you sample at 7.5Hz, it looks like you are producing about 85 bits per sample (I assume a sample is an image frame). That’s too slow (about 640 bps) for most practical applications, but the system is not inherently limited to that rate. The theoretical bit rate is probably around a megabit per second, even sampling at 7.5 frames/sec, but that could not be produced with the present algorithm.
