Noise evaluation algorithm consistency ?

[bitli]

Noise is evaluated and written to a FITS keyword by a few processes (at least IC and debayer).
Are all process using exactly the same algorithm ?
The NoiseEvaluation script uses noiseMRS, is it the same one (there is also a noiseKSIgma on the Image object).
thanks
-- bitli

[bitli]

To answer my own question, I have tried on 3 CFA images. 
The debayer process put the following values for noise estimate:

Code: [Select]

Noise estimates :  8.9240e-005 8.1520e-005 8.7630e-005
Noise estimates :  8.9560e-005 8.2000e-005 8.8350e-005
Noise estimates :  9.0670e-005 8.2920e-005 8.9250e-005
but if I select 'ignore noise' on ImageIntegration to force recalculation I get :

Code: [Select]

Noise estimates :  8.1765e-005 7.4343e-005 7.9577e-005
Noise estimates :  8.2028e-005 7.4770e-005 8.0294e-005
Noise estimates :  8.2950e-005 7.5673e-005 8.1180e-005All use the MRS algorithm.
So it seems that there are differences in implementation, or something that I misunderstood.
Maybe it is safer not to count on the noise evaluation of Debayer, to be sure to have consistent results?

EDIT: As this was posted I did see the light! The images were registered between the debayer and image integration.  Doing an ImageIntegration on the non registered images (just to check the calculation of the noise) I get

Code: [Select]

Noise estimates :  8.9241e-005 8.1517e-005 8.7610e-005
Noise estimates :  8.9560e-005 8.1996e-005 8.8342e-005
Noise estimates :  9.0667e-005 8.2923e-005 8.9249e-005which is very similar to the result of Debayer (the code is not shared, but the algorithms seems to be the same).

For the purpose of weighting I guess that the noise calculated before registration could be good enough, it seem to be transferred to the output image by the star alignment process. For the sake of consistency, it may be better to leave II calculate its noise.

EDIT 2: This was discussed (long time ago) at http://pixinsight.com/forum/index.php?topic=3320.msg22872#msg22872

Sorry for the noise of this post 

-- bitli