LRGB and Narrowband - pre-stretch etc. before combination?

[Amaranthus]

When using LRGB data, what I have done in the past is create the 4 aligned stacks (in DSS or Nebulosity), and then load and integrate these 4 channels in ST before doing any post-processing. That is, all post-processing is done on the combination image in ST.

However, when reading the "Deep Sky Imaging Primer", the author suggested that one should often pre-stretch or even fully post-process each of the channels separately, prior to final colour combining, to ensure that you don't get a given channel dominating. He said this was particularly important for narrowband when Ha can overwhelm the weaker SII and OIII signals. But... the author was focusing on post-processing in Photoshop (side note: as I was reading those chapters, I thought again and again - but this is soooo much simpler, more intuitive and more flexible in ST compared to PS!).

So Ivo, what is your advice on this? I've been accumulating some serious data (14+ hours of integration time) on a couple of object this month (clear skies, no moon, AP party time!), and I want to make sure I treat it optimally!

A second question, in Nebulosity 3, which I now use for all my pre-processing, it is recommended that one does either a 'normalization' or 'histogram matching' of the subs prior to drizzling (I always use drizzling if I can with my current image train, because it is seriously undersampled). I found I get more consistent results with normalization, but I wondered if you had a view as to what was most compatible with ST's "leave it as clean as possible" philosophy? (The drizzle doesn't seem to work well at all without some kind of histogram equalization in Nebulosity).

Thanks Ivo - and I say again, ST is a fantastic bit of work. I talk it up at every opportunity!

[Amaranthus]

Further to this, I've been reading up about how to best create a synthetic Luminance channel. I noted in another thread, Ivo, that you mentioned that ST automatically creates a synthetic Luminance layer, but it wasn't clear if you only meant for OSC DSLR extractions, or also for mono LRGB data, or indeed how you can tell ST about the relative weighting (integration time) of the L vs the RGB data (I could only see how to tell ST the relative R-G-B weightings).

One method for L-LRGB, outside of ST, would be to create the L, R, G and B stacks separately, then create a summed RGB greyscale stack (I'm using equal exposures for R, G & B), and then to an average stack this RGB combination with the Lum stack. But my Lum stack often has a different total integration time to the combo RGB stack, so this method would ignore this critical information.

Needless to say, I'm a bit stuck on the best course of action! What I want to make sure I don't do is throw away valuable luminance data from my R, G & B channels (they're surely worth more than just their chromiance information )

[admin]

Hopefully, this clears things up.
http://www.startools.org/forum/viewtopic.php?f=10&t=589

but it wasn't clear if you only meant for OSC DSLR extractions, or also for mono LRGB data

Indeed, it does this for OSC/DSLR data only.

reading the "Deep Sky Imaging Primer", the author suggested that one should often pre-stretch or even fully post-process each of the channels separately

You can certainly do the same in StarTools if you wish. In that case, process your synthetic L image (in grayscale) separately from an RGB composite. The latter only requires very little processing - just get the colors right. Once you have processed both used the Layer module to Inherit the other image's brightness or color and you have your LRGB composite!

A second question, in Nebulosity 3, which I now use for all my pre-processing, it is recommended that one does either a 'normalization' or 'histogram matching' of the subs prior to drizzling (I always use drizzling if I can with my current image train, because it is seriously undersampled). I found I get more consistent results with normalization, but I wondered if you had a view as to what was most compatible with ST's "leave it as clean as possible" philosophy? (The drizzle doesn't seem to work well at all without some kind of histogram equalization in Nebulosity).

As long as the end result is still linear and all amount of linear signal manipulation is performed the same across all subs, all should be good. I'm not entirely sure what either procedures do and whether new normalization values are established per sub (which would be bad) or per stack (which is fine)...

[Amaranthus]

Thanks Ivo, this is really helpful.

One follow-up question -- for creating the RGB colour stack, would you advise using the LRGB module to load and weight the 3 channels (ignoring the L), or would it be better to do it all via the Layer module?

Regarding what Nebulosity does in its normalization: for histogram matching, as I understand it, you choose a sub which you think has the best/most representative luminance histogram, and then each sub is compressed/stretched to match this histogram. For 'normalize', each sub has their histogram stretched equally across the dynamic range. Here is what the manual says:

Normalization and Histogram Matching

Ideally, all frames taken under the same circumstances of the same target should all
have the same intensity. Often, this is not the case as changes in light level, cloud
cover, etc. can change the intensity from frame to frame. Further, if changes are made
in the capture settings (e.g., different exposure durations), you're certainly going to
have differences in overall image intensity across frames.

If you're doing the Average/Default method of stacking, you need not worry about this
issue unless the changes are really quite severe. If you're using standard-deviation
based stacking, Drizzle, or Colors in Motion, it is a good idea to normalize your images
before stacking. What this will do is to get all of the frames to have roughly the same
brightness by removing differences in the background brightness and scaling across
frames.

There are two methods to normalize the images in Nebulosity.

1) The Normalize entry in the Batch menu will go through all selected frames and
attempt to put them all in a common intensity range by taking care of offset and scaling
differences across frames. After normalization, all frames should have their minimum at
~100 and their maximum at ~65535. Do this before you do any alignment of the frames
(it can be done before or after pre-processing, but you don't want the black borders
surrounding the image that can come in during alignment to throw off the normalization
process.)

2) The Match Histogram entry in the Batch menu will ask you for a reference image and
attempt to match the histograms for a set of target images to this reference image. It’s
performing a more advanced stretch than you’d get in Normalize and can thereby be
more effective. Note, if you’re using a one-shot color camera, do your demosaic
before you try to use Match Histograms.

[admin]

Thanks Ivo, this is really helpful.

One follow-up question -- for creating the RGB colour stack, would you advise using the LRGB module to load and weight the 3 channels (ignoring the L), or would it be better to do it all via the Layer module?

Just use LRGB - don't worry too much about correct weighting as we're mainly interested in getting the correct colors out. The Color module will balance ('weigh') the colors.

Regarding what Nebulosity does in its normalization...

Sounds like this is mostly innocent, unless intensity really is markedly different between frames.