Can I choose to not normalize?

Hi Adam @singding,

No background neutralization is a multiplicative correction on the color channels to get a neutral background.

If you don't want to apply normalization to the frames in your stack, simply set the normalization method to none in 5) .

The background neutralization itsef is not normalization between the frames themselves. You can look at that as normalization between the channels of each color image.

But please realize that disabling normalization is something you really don't want to do. The results will be worse. Thanks to data normalization, you get better data integration and better outlier rejection 😉

If you have data shot at 1 particular session, then disabling normalization will not cause too much harm, but if you combine data of several nights, it really will..

I think, with regards to startools from your other post, that startools is complaining about something having to do with the data being non-linear while the data actually is. I would think that you need to contact startools support about that then. Have you done that already?

It is usefull for me to know what startools actually says about the data, so please share, maybe we can get to the bottom of this then.

Mabula

Thank you Mabula!

Here is what Ivo, creator of StarTools advised:  just curious how to obtain the best stack for both APP and startools 🙂

"

1. As a rule of thumb, apply as little "fancy" stuff as possible. Anything that meddles with the pure data as recorded (e.g. pure photon counts) should be avoided. Every time another program touches the data, however well intentioned, it means that StarTools will yield less and less optimal results - it needs to be able to track signal evolution from as early as possible. 

Hi Adam @singding,

..pure photon counts...

APP gives just that as good as possible, Ivo must mean "relative photon counts", other applications very easily destroy the photon counts in for instance flat-field calibration (having a blue light frame, which turn turns red after calibration...), APP does not 😉 by correcly maintaining ADU counts of all channels.

The recorded ADU counts in your data should be linear to the number of incoming photons. Of course, not all photons are counted, we have the sensor's sensitivity to incoming photons, giving us detected photons.  And we have the gain/ISO used that translate the detected number of photons to an ADU count per pixel. This is what you get in APP's integration results. There is no fiddling/meddling etcetera. APP attempts to integrate the relative recorded photon counts as good as possible, normalization is needed to do just that. We need to subtract the background/sky signal (in normalization) to make the best attempt at integrating the signal/the photons that we are interested in.

If StarTools has a problem with APP data in this regard, then the source of problems most likely (I think) is to be found in what StarTools internally is trying to do/detect/analyse. All processing in APP is based on the assumption that we work on linear data, because it

• gives us statistics that actually  can be compared sensibly
• gives us most freedom in postprocessing the integration result
• gives us a result that contains all information provided by the individual subs that were used in the integration.

As soon as a non-linear stretch/instruction is applied to the data, we basically forfeit these 3 major benefits (not 100%, but it depends on how strong and complicated the stretch is more or less ). We loose a tremendous amount of flexibility, statistics will make no sense at all, and we have compacted information in regions of the data range that will only be recovered if we know the inverse of the non-linear operation, which we usually don't... Perhaps superfluous, but I am not a very big fan of non-linear processing for these reasons, I think Ivo isn't as well... (I do know, that non-linear processing can give very good results as well, only it is much more complicated and requries much more experience. In addition, the non-linear results can always be achieved using the linear route with much less effort, at least in my own experience.)

So, based on these arguments, the fact that StarTools (and thus Ivo) want the highest degree of linear data, is something I can only agree to and recommend 😉 , since that simply is the best starting point 😉 APP really is producing that in it's integration results.

If StarTools has trouble with APP's linear data, I can only assume the problem is in StarTool's algorithms. Off course, I will be happy to communicate with Ivo about this if he is willing to.( I communicated with Ivo in the past actually).

Feel free to copy/paste this or send Ivo the URL to this message.

Kind regards,

Mabula

Thank you. I appreciate all your help, you and Ivo are super helpful 🙂

Mabula, what is the difference between picking a normalization mode (add scale) and clicking checkmark for neutralize background?

If i leave this unchecked does it still normalize, and if I leave normalize to "none" and check neutralize background, what happens? Just curious if you need both, or just one, etc...

Thanks

Adam

Posted by: singding

Mabula, what is the difference between picking a normalization mode (add scale) and clicking checkmark for neutralize background?

If i leave this unchecked does it still normalize, and if I leave normalize to "none" and check neutralize background, what happens? Just curious if you need both, or just one, etc...

Thanks

Adam

Hi Adam @singding,

Neutralizing the background is additional to normalization. Background neutralization is not required, since you need to do background calibration anyway to get a nice sky background. Background neutralization is a method that uses the histogram to neutralize the sky background. Therefore it can surpress weak nebulosity in the sky background. A proper background calibration will restore that.

Normalization however, like I indicated before, is important to get a good result in integration. Image normalization will make the images statistically comparable by correcting each image for the background luminostity and the dispersion of the data relative to this sky background. So by normalizing, you can actually statistically compare the images for noise. Without normalization for dispersion, a noise comparison between iamges is quite useless, so the reported noise values are not related to each other as they should. 

Be aware that comparing nosie between images only makes sense after normalization, something that is often forgotten... If a multiply an image with a factor of 2, the amount of noise in the images is also multiplied by 2 in an aboslute sense. By normalizing, the noise is actually comparable between images.

Therefor if you compare images for noise by simply running them into a noise evalution script per image (which a lot of users seem to do to compare them), then the reported values really can't be compared wisely.

Mabula

Thank you Mabula! I appreciate it!

Just to make this absolutely clear: When I process monochrome images, there will be no difference checking "neutralize background" in 5) or not. Even when it stays checked nothing will happen, because there are no separate color channels within the file. Is that right?

Posted by: schurig

Just to make this absolutely clear: When I process monochrome images, there will be no difference checking "neutralize background" in 5) or not. Even when it stays checked nothing will happen, because there are no separate color channels within the file. Is that right?

That's totally right @schurig 😉

The neutralize background function only applies to RGB /multi-channel data.

Mabula