New Ha Combine Technique

[Jack Harvey]

I have been experimenting with Ha combining as you know from the previous thread.  Well I have developed a new technique that I think has some promise.  I am ready also for everyone to shoot holes in it<G>.

1.  Process the Mean Ha frame with decon, wavelets or what ever you deem to get a good Ha image.  Then mask the image and use ACDNR to smooth the periphery of the Ha nebula, somewhat aggressively.  Adjust mid point with Histogram (see step 2).

2.  Extract the Luminance from a processed RGB of the image using the extract channels, with L checked and a and b unchecked on the CIE button.  You can throw this away as you will not use it.  I did look at the Lum to adjust the histogram for the above Ha to be close to that of the Lum frame.

3.  Now extract the R,G, and B frames form the RGB using the Extract Channels but this time the RGB button.

4.  In LRGB Combination tool put the Ha frame in the Lum slot and the RGBs in the correct slot.  Add a slight amount of saturation (lower the sat slider 15-20%) and start with about 50% of Luminance contribution using the slider in the upper right.  You can run several diffferent percentages of the "Lum" which is really Ha and see how much is best by comparing the output images.

5.  Make final adjustments with histogram, etc.

6.  Here is NGC 3852 I did by this method with 65% Ha contribution.  I think it compares favorably (if not better) to the one I have in the gallery that was produced with the more intensive steps that are discussed in that thread.  http://pixinsight.com/forum/viewtopic.php?t=526

  I am open for comments and certainly critiques.  Thanks

[Juan Conejero]

Jack,

This is an improvement, and your result clearly shows it. Your step 2 is clever because it adapts the new Ha luminance to the existing RGB data, which is very important to avoid "orphan" chrominances and luminances

The drawback is the same as with all methods that mix narrowband data with RGB: noise. The Ha image isn't used at its full potential, since it's mixed with the RGB luminance, which has a much lower SNR.

This is the strong point of Vicent's method: he puts the whole Ha image into the final composite without contaminating it with noise from RGB. His method is complex, however, and still has to be perfectioned (and implemented as a tool, because applying it manually is a crazy thing).

Having said that, I think you're doing it the right way with the LRGBCombination tool. You'll find this tool is completely free from the infamous "salmon-pink" problems and the like, so common in other implementations. This is because we use truly rigorous color space transformations. There is no hue change at all with LRGB in PI, as long as you keep your data within the limits of the RGB space (this will normally be the case, unless you apply a really brutal saturation change).

Well done, and nice result

[Jack Harvey]

Juan   Thanks for taking a look and the critique.  I like Vincents method but as you point out if you do it manually it is very time intensive and complex.  I cannot wait until he has it perfected in the form of a tool.  The nice thing about this method is it is quick, fairly simple and gives good stars and colors.  Of course I have only done a couple of images so it will be interesting to see what results others get.

[Jack Harvey]

Here is a HARGB of Herschels Ray done with the quick and easy new technique which essentially extracts the luminance from an RGB and substitutes an Ha frame in place of the now discarded luminance.  Again I think this image compares favorably with the Herschels Ray image done earlier using the method Vincent has proposed.  http://pixinsight.com/forum/viewtopic.php?t=532.

[Jack Harvey]

Here is NGC 346 done by an old Ha combine method of mixing Ha with Red and then combining to get a HaRRGB Image.  Compare to the new technique of substituting a Ha frame for the Luminance frame that is extracted from the RGB.

Traditional HARRGB Image


New Ha for Luminance Substitution Image

[LD]

Jack,
First, thanks for the email steering me to your thread. Second, the results really do speak for themselves; that final example is quite dramatic. I've got to try it.
But I am confused on two points: when you extract the luminance in the CIE space with a and b unchecked, are you left with an image less the luminance? Just not sure I followed that step. My second question is, how is this method different than creating an RGB image and then just using straight Ha as the luminance via LRGB combination, which is my old and unsatisfactory method. (I'm favoring the HaRHaRHaGHaB method you pointed me to a few months ago.) I can certainly see the results versus your HaRRGB, the improvement being quite apparent. And I know I'm missing something (my usual MO :oops: ), but I just can't figure why it's not the same as creating an HaRGB in the old fashioned method. :?
Then again, sure seems easy and seems to work great!
Thanks again,
Larry Durst

[Jack Harvey]

LD  Yes you take a processed RGB and then extract the Luminance using the extract channels as you understand.  This leaves the RGB with the Lum channel gone.  Then take a processed Ha image and stretch it with histogram mid point to fairly closely match the extracted Lum.  At this point you can also do some other stuff to the Ha like ACDNR if necessary.  Now you can throw away the Lum.

Take the same RGB from which you extracted the Lum and now use the extract channels set to RGB and break the RGB into R,G & B frames.  

Put the Ha in the Lum slot of the LRGB Combine tool and the R,G, & B in the appropriate slots.  Increase saturation by moving the sat slider some to the left and move the Lum slider to give somewhere between 50 and 90% Lum (Now actually Ha) contribution..  Hitr the Global process button and you now have a new image, a HARGB.  I usually do a 50%, then move the slider to get a 65% and then again to get a 80%.  I visually select the best result and finish off processing with what ever tools suit your fancy (Histo, Curves, etc)

Another option is to use Pixel Math and add the RGB and HaRGB at a 1:3 ratio or so if you want more saturation to the final product.

Now the hard question, I do not know why but this method solves the star issue and just works better than adding Ha to a RGB (which of course has the Lum component).  Juan in his note mentioned something about the importance of using the LRGB combine tool to do this, so perhaps that is p[art of the explanation.

What I know is I have used it on about 5 images and all are either as good (usually better) than my previous efforts<G>.  An added bonus is it is quick!  Let me know your result.

Summary


RGB>>>extract Lum>>>extract R,G,B channels>>>put Ha,R,G,B in LRGB combine>>>final Processing

Ha>>>match extracted Lum in midpoint visually (now throw Lum away) and use in LRGB Combine

[Jack Harvey]

This is an improvement, and your result clearly shows it. Your step 2 is clever because it adapts the new Ha luminance to the existing RGB data, which is very important to avoid "orphan" chrominances and luminances Smile

 Juan


LD  This may be a clue to why this works, but Juan may need to elaborate more<G>.

[Juan Conejero]

Hi Jack and Larry,

First I must say that I'm glad to see experimental works like this one here. This is our goal with PixInsight: experiment and explore new paths. This is where the real fun is

"orphan" chrominances and luminances

I mean that if you change the luminance of a RGB image, you must ensure that the new luminance provides enough support for the chrominance, and vice versa.

If your luminance doesn't give support to the existing chrominance, you'll get a dark and noisy image. If your chrominance doesn't support the new luminance, you'll get an unsaturated result.

This is why we included the luminance and saturation transfer functions in the LRGBCombination tool. By fine tuning them you can achieve an optimal mutual adaptation between luminance and chrominance.

In your schema you adapt the Ha to work as a new luminance for the RGB data *before* the LRGB combination process (step 2). This is a good idea for the reasons above.

I think one answer to the question "why this works" is that you have RGB data of exceptional quality. Your broadband red image seems to provide a rather good chrominance support to the Ha image that is working as luminance. If your red channels were poor (noisier, weaker), the obtained results could be disappointing.

Another obvious answer is that you are throwing away the implicit luminance of the RGB data, which is usually poor, and replacing it with a higher SNR luminance, the Ha image in this case. This of course improves the SNR of the result. This happens also when you do a regular LRGB combination, but a narrowband image as Ha has more signal and less noise than a broadband luminance.

A drawback: this works for emission nebulae, but not for OIII emission, or reflection. In these cases, the Ha image cannot work as a luminance because these objects aren't red at all. The fact that you are mixing Ha and L to form the resulting luminance partially solves this problem, but at the cost of reinserting part of the noise from the RGB data into the final (Ha+L)RGB image.

Finally, the stars are turning out well for the same reason: you are reinserting part of the RGB luminance in the result, which gives enough support to the stars.

I think this is an interesting approach. It's more likely to work well with high-quality RGB data, as is your case, and Ha emission objects. It's easy and can be a good alternative to the usual (Ha+R)RGB method. You should experiment with more images, including bad RGB data and different objects, to see what happens.

[Jack Harvey]

Juan thanks for the information.  I don't have any "bad" RGB data to exsperiment with<G>.  Also you will note I have extended this idea a little further in using it for a LRGB rather than HaRGB.  That thread has an imge of the extracted Luminance and certainly shows what you said about the noise.

[LD]

Did someone mention "bad" RGB data? I've got plenty! VBG
Larry
BTW Juan, I do agree about experimenting. The true beauty of PI is that it does allow so many creative approaches. It really brings out the problem/solution side of processing and opens up a whole new side of imaging that I had thought might be out of reach of non-programmers like myself.

[vicent_peris]

Hi all,

some quick comments, as I have no time!

About my method: the resulting image is only a combined raw image. I mean that it's the base to start to work with the data. It has some advantages: the HaRGB composite has not the noise of the RGB data and it has some enhancement of nebular structures. BUT, of course, you can process separately the H-alpha and the RGB images and combine them in the last steps. In fact, I highly recommend this!

It's like having a broadband LRGB image. You can process separately the luminance image and the RGB data. And after the processing, you make a new combination. Working like this is highly recommendable: you will exploit at the maximum the luminance and the chrominance of the image.

So, Jack's methods brings out this workflow. Jack, think that, if you previously have a better RGB data, along with the enhanced nebulas, the results presented here would be even better!

I think that this is not truly a technique, but a very good and highly recommendable workflow. Good work, Jack.  :wink:



Vicent.