[01.08.03.1115 Ripley x64] Noise Generator process

[Bart_van_der_Wolf]

Hi All,

I'm trying to understand the (undocumented) Noise Generator process, because the Poisson noise statistics are completely different from the Normal distributed Gaussian noise, even for high levels of exposure. At high signal levels (which would take long to calculate for Poisson noise) the noise is expected to approximate the Square root of the signal.

It seems like the standard deviation of the Poisson noise is several magnitudes lower than expected, even with the Amplitude control maxed out at 1.0.

The post that announced the new bugfix stated:

* NoiseGenerator process: The Poisson distribution option was not working correctly (the generated noise was actually Gaussian). Poisson noise is now generated correctly. In addition, noise generation now tends to preserve the expected value of the target image.

Maybe something went wrong with the last mentioned adjustment (marked in bold)?

Cheers,
Bart

[Juan Conejero]

Hi Bart,

The NoiseGenerator process computes for each pixel a (pseudo-)random deviate from a Poisson distribution with mean equal to the current pixel value, assuming a 16-bit linear "camera". So if v is the current pixel value in the [0,1] range,

v' = Poisson( v*65535/a )/65535*a,

where a is the amplitude parameter, Poisson( x ) computes an integer pseudo-random deviate from a Poisson distribution with expected value x, and v' is the resulting noisy pixel value in [0,1]. So the amplitude parameter acts in this case as a "signal limiter" by reducing the effective bit depth of the simulated acquisition device.

Note that this procedure only makes sense, as a simulation of the physical photon counting process, for linear images. If you apply this tool to a nonlinear (stretched) image to generate Poisson noise, the result will have no physical meaning with respect to the existing pixel data.

[H.Geske]

Hi All,

Yesterday i have finally updated Pixinsight from Version 1087 to 1185.

Usually i have used the Noise Generator and Poisson Filter for flatening the background of Moon and Planetary images, wich worked always fine with Version 1087.
Now with Version 1185 it has no significant effect to the picture.

Is this made by purpose?

After i have read this Thread, i have tried it on a linear image. There i can see a little effect.