subtract_image / subtract_threshold implementation iso conform?

[HTuennermann]

A colleague of mine brought up that the clipping to zero done in these functions is not mentioned in the iso standard. Therefore this wouldn't be completely iso conform. Instead, negative numbers should be used for the next steps. Would you be interested in changing this?

[scottprahl]

I would be happy to consider a pull request. Please include the relevant text from the standard (I'm in academia and don't have access to the standard.)

[miromarszal]

I don't have access to the standard either, but I'm pretty sure that clipping the noise to zero is incorrect. The background is not zero mean after such operation, hence there is an average offset — precisely what you are trying to avoid by subtracting the background in the first place. The change of noise statistics is less of a concern but something to keep in mind too.

[HTuennermann]

"As a result of a proper background subtraction there must exist
negative noise values in the corrected power density distribution. These
negative values have to be included in the further evaluation in order
to allow compensation of positive noise amplitudes." (ISO 11146-3 clause
3.1)

however unfortunately this is more tricky to change than i first thought as the current examples become unstable then (hence why I did not send any pull request yet)

[scottprahl]

I think that the way to go is to add an option to functions for clipping ['strict', 'non-negative'] or something similar. We should support the standard even though some of the test cases fail (often these are extreme cases that no serious metrologist would ever make)

[scottprahl]

Thanks for the suggestion. The newly released version allow negative pixel values after subtraction. If the background is reasonable, then this method works fine. However, for many of the student images, the background was very large and to find the beam it is necessary to pass the option iso_noise=False to the function.