Can somebody explain these strange process console statistics?

[Geoff]

I've been looking at the ImageIntegration statistics provided in the process console and the results seemed unreasonable (v high SNR for example).  As a sanity check, I took three identical copies of an image and ran them through ImageIntegration.  The process console indicated a reference and median noise reduction of 1.15.  The integrated image and the original image were (as expected) identical.  So how can there be a noise reduction? Also, the claimed SNR is around 600--on a 10 minute sub!?

[Juan Conejero]

Hi Geoff,

The SNR estimate is computed as Equation 12 in ImageIntegration documentation. Note that this is not an estimate of SNR increment, which is what you probably expected here. This SNR estimate is given for informational purposes, although it is not a robust estimate for the reasons noted in the documentation, so it must be used with great care for comparisons.

The effective noise reduction estimate (ENR) is given by Equations 40 and 41. If you have selected the same exact image three or more times in ImageIntegration, the ENR estimate should be exactly one. I have just made a test and ENR = 1.0 as expected.

If you get an ENR estimate of 1.15, then there are different input images. Can you double check this?

[Geoff]

Hi Juan
I have just redone the test with a different file.  There is a similar anomaly, as the attached file shows.

[msmythers]

Geoff

I tried with one of my DSLR frames and got the expected result. Have you tried clearing and purging the cache in the ImageIntegration preferences?


Mike

[Geoff]

Hi Mike.  I presume you mean unchecking the box that says "use file cache".  Yes I have tried this.

[Geoff]

The effective noise reduction estimate (ENR) is given by Equations 40 and 41. If you have selected the same exact image three or more times in ImageIntegration, the ENR estimate should be exactly one. I have just made a test and ENR = 1.0 as expected.

If you get an ENR estimate of 1.15, then there are different input images. Can you double check this?

Thanks Juan.  There is really something weird going on here.  If I take 3 copies of an uncalibrated image, I get the expected value of 1.0000.  If the same image is calibrated and three copies integrated I get a value of 1.0002. If the same image is calibrated and registered (against my best sub) I get a value of 1.14...

This occurs on two different computers, an Asus laptop running Win 7 with the penultimate version of PixInsight and a Dell desktop running Win 10 with the latest version of PixInsight.
Geoff

[msmythers]

Geoff

I'm referring to the preferences for the tool, the little wrench icon.

Mike

[Geoff]

Thanks Mike.  Just tried that now.  Same result

[Geoff]

Here is he link to a file showing the anomalous behaviour (on my computers)
http://endor.uv.es/files/data/public/2f9832.php

[Geoff]

It seems that at some stage during calibration and registration, a noise value has been added in the file header.  This seems to be the case with both the BatchPreprocessing script and if it is done the long way.  Maybe using the same file causes a glitch.

Here are some figures:
Noise estimate in file header 1.5190e-4.  Noise estimate in process console1.3287e-4. So the "improvement" factor is 1.5190/1.3287=1.1432, exactly as reported on the console.
Geoff