General advice

Comments

• BlockHead

  June 2018 edited June 2018

  Hi Alan,

  
  

  Well.. there aren't enough users yet on the forum (I haven't even really "announced" my new Pixinsight material- which I will likely do this week). So here is a quick summary of some things that might be helpful.

  
  

  As you review the videos.. very soon you will likely learn the answers in your own way!

  
  

  1. Calibration, Alignment, data rejection and stacking- there isn't a difference in capability between CCDStack and PI here. These are simply numerical operations. PI does offer different/more options in terms of normalization and rejection. The one area I am still trying to figure out is the weighting scheme of PI. It doesn't have the same (simple) method that CCDStack uses. This is something of a techy/nuanced thing I am still working on. Indeed, with the introduction of "large scale rejection"- you can now do Selective Rejection which I LOVE and I found a PI method. Please see this section.

  
  

  2. Deconvolution in PI is multistep. The creation of a PSF is basically the same kind of thing in both programs- although in PI there are extra degrees of freedom. The inclusion of an image that protects bright stars and features from the ravages of deconvolution is also good. If you are familiar with my Photoshop technique I would have to protect bright stars with a mask. This is somewhat synonymous- and actually a bit easier to do. Then there is another global application of deringing. I demonstrate how to control this stuff and attain good results in PI. Once you see my method- you will find you can obtain the same kind of result that CCDStack obtains. There is an extra more "advanced" feature of PI called regularization which is something like the noise floor control that is automatic in CCDStack. You can control it in PI.

  
  

  3. DDP-like effects in PI. There are a number of tools- but what I demonstrate most clearly is the the combination of deconvolution and HDRMT (a process you will learn a lot about!) is super powerful- DDP is actually defined in a very particular way (a family of non-linear curves). HDRMT is even more spacious because you can operate in different degrees of non-linear application on *different scales*. That is really powerful. PI gets a big big check for this cool tool not found anywhere else that I know of.

  
  

  4. Noise reduction in PI is as good if not better than Photoshop due to the number of different ways to do it. (no less than 5). TGVDenoise itself is probably the main workhouse. However MLT in combination with is should take care of quite a few noise patterns. 

  
  

  5. Mask generation. Masks are no or less powerful in PI or any other program... they work in the same way. Creating some masks in Photoshop is a little more organic- mostly because you use them in combination with layers. PixInisight does not use layers (in the same realtime way). So the state of mask at the time you apply a process is all that matters. The creation of some masks takes more steps in PixInsight. However there are other kinds of masks that would make your head explode in Photoshop to create..but you can do it in a few clicks in PixInsight.  For example, it is possible to capture all stars that are larger or smaller than a particular size in a single step. This isn't easily done in Photoshop- because selection of stuff is usually based on brightness... not structure size. 

  
  

  6. Cosmetic Corrections. This is a term that means something in PixInsight- and it is a global process.  If however, you are asking about content aware types of corrections (say a healing brush)- this isn't in PI. So there are different methods of interacting with images in PI. The solutions depend on the problem- so it is hard to generalize all of the methods. The substitution of pixels by using MultiLinearTransform or MultiMedianTransformation are unique to Pixinsight- and quite magical. In fact my method of de-emphasizing stars relies on this functionality. It is an innovation I am so excited to share with others... and see what they think!

  
  

  7. DBE as you know is worth the price of the software. What an improvement this can make the quality of images.. it is certainly a PI crown jewel.

  
  

  So when you use a PI function and it gives you a result that you find more pleasing- I would argue this is a different result- not a "better" one. PI uses different algorithms and some of them are unique. Sometimes a result is only possible through multiple steps. This result may not be available in other software- but still requires effort in PI. So its state of "betterness" is related to a user's tolerance of whatever "easy" is defined as. I am happy to jump through hoops to get a result I cannot achieve elsewhere. Of course it is silly to do multiple steps on something that is accomplished in few steps in another program.

  
  

  -Blockhead

  (I am embracing the bullying of my youth)

  
  
• Alan

  June 2018

  Adam

  
  

  Many thanks for your detailed response. 

  
  

  Yes, I'm slowly going through the Pixinsight Fundamental Tutorials and so far I've been very impressed, having already discovered quite a few useful techniques. I obviously have more to discover !

  
  

  To respond on your specific points:

  
  

  1. Calibration, Alignment, data rejection and stacking - OK. I understand, however, for the time being I will stick with CCDstack since I personally find it much easier to do the large scale rejection by simply "painting" with the pen on the objects to be rejected. I also don't understand the PI subframe weighting scheme since I would have thought that the main item you are interested in within the subframe should be how the weighting scheme is mainly calculated (as can be achieved in CCDstack) as opposed to calculating the weighting over the entire subframe which PI appears to do.

  
  

  2. Deconvolution - OK, I shall experiment with this.

  
  

  3. DDP like effects - Yes, this sounds interesting, I've found HDRMT very powerful in PI and personally I find this much better that the PS HDR processing nearest equivalent. 

  
  

  I'd like to improve the way that I initially generate my initial Lum which currently is via a DDP permanent stretch in CCDstack, making sure that the object of interest doesn't have pixels brighter than 200 (ish), so that you can achieve colour when in PS and blended with the RGB data.    Whilst this works very well for the object, if the stars are brighter than then the object, then it can lead to excessive star bloat and lack of star colour. Currently, I overcome the problem but doing two types of DDP stretches:one optimized for the the stars and one of the object and then blending the results via PS. I'm sure there is a better way of do this in PI.   

  
  

  4. Noise reduction - I generally find that PS noise reduction on quite bright objects seems to work quite well but it is not so good at dim objects (eg background sky) - by contrast PI TGV seems much better at obtaining a smoother background.I haven't tried the MLT noise reduction options but I will !

  
  

  5. Mask Generation - I understand what you are saying but for now I shall stick to PS for mask generation. 

  
  

  6. Cosmetic Corrections - I was referring to the healing brush in PS. I don't really understand the PI way of doing this but again I'm willing to learn.

  
  

  7. DBE - yes agreed. PI offers the best method here - it was the reason that I bought the software !

  
  

  Alan

  
  

  
  

  
  

  
  

  
  

  
  

  
  

  
  

  
  
• BlockHead

  July 2018

  Thanks for the response Alan. 

  I understand where you are coming from...because I came from there!

  I do note, concerning a few of your points- that there are some sections you may not have yet reviewed that will inform some of your comments. I would list some of the sections that address them..but I think you will get there (and maybe write again).

  
  

  Concerning the weighting scheme of Image Integration...I KNOW- I have been delaying recording this lesson until I gain a full understanding of getting the weights I expect. 

  
  

  Concerning the selective rejection- I assume you saw my section on this? I actually do prefer the PI method as it is now identical to CCDStack (except you have to make an image to paint values from)- but with the added benefit of being able to copy the painted areas to other images!! It used to be I would have to selectively reject lots and lots of dust donuts... but if they are in the same place-now I just make an image container and viola! all of the images are painted. Very cool. I am biased though...I am pretty "proud" I brought this technique to PI.

  
  

  -the Blockhead
• Alan

  March 2019

  Hi Adam

  
  

  I thought I'd update this thread after 8 months.

  
  

  I'm still using PS and CCDstack but I'm using PI even more !

  
  

  Whilst I still use CCDstack for calibration I now use Pixinsight for alignment and stacking. The reason for my transition was two fold. Firstly, I came to the conclusion that Pixinsight's large scale rejection features gives a better rejection of satellite/plane trails. Whilst you can get more or less the same result in CCDstack via use of the freehand tool - I now prefer the PI solution. I was also impressed by the Local Normalization function in PI which does seem very effective in compensating for changing sky conditions (eg light clouds) between subs. The second reason, is that I really like the PI noise reduction script Mure Denoise, however, this is only effective if Mure Denoise knows the alignment algorithm used for the alignment process. As far as I can tell, Mure Denoise would only be effective on CCDstack stack outputs which have used the Nearest Neighbour registration algorithm. 

  
  

  Alan