BXT after star removal?

Scott Cunningham

I noticed in your workflow examples that you will remove stars (SXT) before doing BXT on each of the narrowband channel images. Doing it that way, you have to uncheck Automatic PSF and figure out where to put the slider for PSF Diameter in pixels. Is that better than running BXT before removing the stars?  If this is the better way, then how exactly do you determine how to set the PSF diameter? I didn't see that in any of the videos. You offhandedly remarked to "set it appropriately." I have no idea what is appropriate or how to determine that. Does it have to be set exactly to some value? Can you please elaborate on the proper workflow order here between the star removal and deconvolution steps, and how to determine where to set the PSF Diameter slider? Thanks!

BlockHead

" I didn't see that in any of the videos. You offhandedly remarked to "set it appropriately." I have no idea what is appropriate or how to determine that"

I just want to say this is one of the many examples where Horizons and content within it assumes some knowledge already- specifically ideas and concepts that I explain in Fundamentals. So... the way you put it I feel indicates an incompleteness. However, I can't explain everything every time in every video and I clearly state that Fundamentals is a good prerequisite for Horizons.

I did even make a video on this very topic:

https://www.adamblockstudios.com/articles/psfs-blurxterminator

At the moment, I cannot remember the reason I chose to do this- I suspect it is because the stars look so different in each narrowband image it may not be optimal. But assuming I had good reason to make the suggestion, the answer is you set the PSF based on the measured FWHM of stars in the data. It doesn't have to be terribly exactly. Are your stars 1 pixel in size (undersampled), two pixels? three pixels or more? If you want to measure it precisely you can use dynamic PSF or Hartmut Bornemann's script. 

The behavior is, the larger the PSF the more "sharpening" like effect you get with BXT. 

-the Blockhead

Scott Cunningham

After having searched through all the videos, I am unable to find a section that describes why you perform BXT where you do, in relation to other workflow processes. You narrowband workflow doc (PowerPoint doc) has BXT after SXT, Linear Fit, ChannelCombination, etc., as is the case in many examples. I used to run BXT on each of the WBPP output files for the narrowband channels with the stars, but following your workflow is very different. Is there a reason I should be doing it this way? I seem to get great results doing BXT first, but haven't had the patience to go through all the steps doing both ways and compare. But I suspect the difference would be too small to be of concern.

Thank you for your patience. With literally hundreds of videos that I am going through and trying to remember everything, I sometimes need a question addressed before I can continue productively, or else little of the following material makes sense to me, at least in the sense of the order in which you are doing it. You often mention that you don't like to provide workflows. As a beginner, without a workflow to follow along with, I tend to get confused by the order in which you do things. It would be extremely helpful to someone without 30 years of image processing experience to be able to follow along with at least some sort of overall map of where we are starting, where we are going, and why that particular order, before you dive into the details of the processes. My engineering brain wants to start from a 60,000 foot view and then drill down to the particulars, in the correct order.

This is particularly the case with where is it best to perform BXT in the workflow, and why not do it at the very start, right after WBPP? That's pretty much where all the other YouTube "experts" perform BXT. I have learned so much about each of the particular process you describe in both Fundamentals and Horizons, but I get stuck when I look at my own data and try to decide what is the best order in which to do those processes. I have created various workflow diagrams for many of your examples so I can recreate what you did without having to start and pause the videos a thousand times while I process my own data while watching you. Sometimes there seems to be somewhat contradicting strategies between examples.

Should I try to process my own stuff as I am watching and learning from your videos? Or watch them all first (100s of hours) and then try to process my own stuff?

I do mostly monochrome narrowband with good equipment. I do many 10 of hours of imaging and get silky smooth output from WBPP on my SHO channels, which I think I have mastered fairly well.

Here are my latest images processed using your techniques from Fasttrack and Horizons. Are these acceptable?

https://www.astrobin.com/users/stacaz99/

For me is it very helpful to have a workflow to follow, with conditional branching for things I want to accomplish with the final image. Posting these for each type of DSO object type would be greatly appreciated.

I apologize for my blockheadedness and perhaps a bit of impatience. I have a thousand more questions that I will slowly post over the coming months as I try to be a better student of all this.

Thanks!

BlockHead

Here is the thing. You are making my case for why workflows are not the answer in the long run. You want an order. But I say the ordering depends and may change. So I say one order and then in another video do something else it will appear if I am not being consistent. In fact I do exactly this. In the Helix example I use BXT on the mono channels before making a color image of them.

One overarching idea with BXT is that it looks at the relationship between the color channels. So if you have a luminance (or a channel you are using as the luminance) you apply BXT to this. BXT is trained on broadband data. So the choices are to do BXT on the individual mono channels OR do it on the starless color images. It is weird to me to do it on on NB color images with stars. That does not seem right. 

In the Wizard example I think I wasn't really considering it a big deal. Not much sharpening was going to happen (only in the small bright area) because most of the object is so faint. It wouldn't matter if I did it on the individual channels (after linear fit) or in the color version. Perhaps I was being lazy. I suspect what I was really thinking was to keep things simple and do the fewest operations (just on the color instead of the individual grayscale images). 

 In all cases you would want to use BXT after any linear fitting. That much is pretty solid. So I disagree with all of the experts. I believe you should do BXT on the scaled image. I think you can see why- if you do not- you potentially will not be deconvolving the weaker channel as much. Yes the weaker channel will be noisier after being scaled- but this is why you need to take lots and lots of exposures for NB. 

When you do BXT with respect to SXT is really quite minor. On starless images, you will use a fixed PSF which is fine for all but the pickiest deconvolvers. Otherwise, doing it on the starred grayscale  (after linear fit) data is fine. 

"Should I try to process my own stuff as I am watching and learning from your videos? Or watch them all first (100s of hours) and then try to process my own stuff? "

Yes and No. I do think the watching all of Fundamentals before Horizons addresses many of the Horizons only-members questions that I get. But no, the way to approach learning from my videos is that I show you everything I do. (AND I also show you my results... ) I am not a random number generator and I am somewhat consistent. As a viewer you should pay attention and look for the patterns. See what I do every time...and what I do not do of course. Take notes. And lastly you cannot get away from practice. There are no free lunches. It will take some degree of repetition- and your idea of comparing one way versus another is a good one given time.

So you began with a question about ordering- but a better approach (which you mention later) is to ask about the contradiction. "Adam...why did you do BXT on the color image in the Wizard and on the monochannels for the Helix?" I have answered it in this case- it wasn't critical and I think my thoughts were on other more important ideas. BUT- a good takeaway from this is the linear fit followed by BXT...that *is* important. (I am pretty sure I mentioned this.)

Just ask the questions here...and I will respond.

-the Blockhead