Getting the Toes of the Histograms to Begin at the Same Place

[dgbarar]

Hi,

When I first began astro image processing in the 2001 what I learned to do was to get the toes (left side of the histogram to line up).  This practice was referred to by Ron Wodaski as making a "bias" adjustment.

However, with Pixinsight that does not seem to happen.  I typically perform the following:

1) Background neutralization.  Develop a preview in a dark area of the image.  Carefully go over the background area preview with a mouse and determine what the true background level.  Set that as the maximum (plus a little) in the background tool.
2) Dynamic Background Extraction.
3) Color Calibration.  Again develop a preview of a background area.  Carefully go over the the preview to determine the true background levels.  Set the minimum in the white reference and the maximum in the background reference to this color calibration tool.

Yet, when I am done the black point for all three color channels are never the same.  This causes problems when on subsequent stretches because the channels stretch differently.  I have attached a screen shots to show the impact.

Perhaps I am being to much of a perfectionist.  Or, am I doing a very poor job of background neutralization and color calibration.  I would appreciate the insight of others.

Cheers,

Don

[Phil Leigh]

You should do DBE first, then BD, then CC...

[dgbarar]

Hi Phil,

I'll give it a try.

Don

[Phil Leigh]

Sorry I meant to say:
1) dynamic crop to remove any (hard to see) stacking artefacts at the edges of the frame - always do this or DBE may not work as well as it should!
2) DBE
3) Background Neutralisation
4) Colour Calibration

[Harry page]

Hi

Are you rgb imaging as your bubble is not red 

CC is for RGB Imaging

Harry

[dgbarar]

Hi Phil and Harry,

Message to Phil:  Yes, I always may sure to crop the stacked images as the first step prior to any processing. As you pointed out this is necessary to eliminate any "edges" that are not fully represented by all three channels.

Message to Harry: This was a narrow band image.  You are correct, there is no reason to go through a color calibration step on a narrow band image.  Let me try and see how that works.

Don Barar

[dgbarar]

Hi Phil and Harry,

Gave the following a try:

1) DBE.  Most of the points were on the left side of the image and on the corners of the right side.  I was careful not to include any nebula regions.  A total of 44 sample locations.  I also expanded the samples to size 15.
2) Background neutralization.  The reference area was the preview on the left.  Background areas in the preview had readings on each of the three channels from 0.000032 - 0.000043.  I set the upper limit for the reference to 0.00005.  I also tried various reference areas on the left side and different upper limits without much change in the results.
3) HT.  Tried this several ways with STF locked and unlocked for the channels before transferring to HT.   Did not seem to make much difference in how well the histogram "toes" lined up.  Attached is the final result.  The image was over stretched for illustration purposes.

I think the methodology of perfuming a DBE first is an improvement.  Yet, I still do not get a perfect line up of the toes.  The other thing that I noticed is that the green channel (15% Ha and 85% OIII) is shaped significantly different.  The toe begins latter and overall has a much steeper slope at the beginning.  I believe this to be a problem as the green channel histogram stretches differently.  So what should I conclude:

a) Need a better DBE technique.  Do a better job selecting samples and setting parameters.
b) Need a better BN technique. Do a better job selecting the reference area/upper limit.
c) This is as good as to be expected.  Stop trying to make this perfect.

Don Barar

[Phil Leigh]

Well, my own perspective on this is that there is no reason for the histograms to line up exactly in position or shape... unless the images are of a white reference card under pure white light and taken through a Bayer Matrix on a DSLR or OSC and then debayered (and no filters are used).

If you want a sure fire way to FORCE them to line up, do this:

1) split the RGB into R, G and B mono images using Channel Management - Channel Extraction
2) pick one as a reference and use Linear Fit on the other two
3) Use channel Combination to recreate an RGB image.

[dgbarar]

Hi Phil,

This sounded like a good idea so I thought that I would give it a try.  I decomposed the image as you suggested and did a linear fit on the red and blue channels using the green channel as the basis.  The color channels were then recombined.  Attached is the result.  As you can see this had no impact on getting the toes of the histograms to line up.  Good thought though.

Don Barar

[jvaart]

How about the method described in this video? https://www.youtube.com/watch?v=1C_YMhLDlEc

[NGC7789]

...As you can see this had no impact on getting the toes of the histograms to line up.

That's not how I see it. If I compare this histogram to your original one at the top of the thread the toe is indeed lined up. The red channel also appears overall hotter lining up with the blue channel. It's true that the shape of the curves is different and they don't rise from the shadows at the same rate but I don't understand how that can be a goal of color calibration as the shape of the curves is data dependent. Nothing says that a color balanced image will have all channels rise from the shadows at the same rate (if that's what you are claiming.)

[dgbarar]

Hi NGC,

I would agree that the toes of the red and blue channels using the decomposition method line fairly well and the methodology does seem to be an improvement. 

The problem appears to be the green channel where the histogram begins to the significantly (at least I think it does) to the right of the blue and red channels.  I think this is a problem.  If the toes of all three channels do not begin in the same place how can the background be neutral?  Which is what I thought was the purpose of background neutralization.

But then again, maybe I am wrong and being overly critical.  My goal in posting was to find out the thoughts of others.

Cheers,

Don Barar

[dgbarar]

Hi JVAART,

I watched the video as you suggested.  When the poster was completed with his processing it looks as though the red and green channels lined up vary well.  However, the blue channel histogram had a tail that began significantly to the left of the red and green channels.  While this video was done with an older version of Pixinsight, the concepts are probable still the same.  Do you know if there are additional videos from this author that address additional processing?

Cheers,

Don Barar

[jerryyyyy]

One comment.  I do not think you have to do the Background Neutralization any more after the newer versions of DBE.  I hope the extra step is not screwing up your processing. 

[NGC7789]

The problem appears to be the green channel where the histogram begins to the significantly (at least I think it does) to the right of the blue and red channels.

Maybe the problem there is that you used green as the reference. This obviously means the green channel will not be changed. Looking again at your original histogram it look like blue is the hottest channel. Maybe this should be used as the reference for the linear fit.