Starmask behavior differs between preview and full image

[SeaLandSkyPhoto]

Hi,

I was playing around with starmask using a preview dragged to the workspace (for speed in experimentation).  I came up with settings that I liked and then applied the same process to the full size image and got radically different results.  I'm confused. 

BB_NL_Stars_L is the full images (non-linear).  I created Preview02 and then dragged it to the workspace (BB_NL_Stars_L_Preview021).  I then applied the starmask process to BB_NL_Stars_L_Preview021 and got star_mask3.  I then applied the same starmask process to BB_NL_Stars_L and got star_mask4.  None of the stars showing in star_mask3 show up in star_mask4.

Thanx,

Paul Howard

[lucchett]

Hi,
from my knowledge the starmask tool algorithm samples the entire image, so the preview is not viable.

I remeber an old post from Juan on this.
Best
Andrea

[SeaLandSkyPhoto]

Hi,

Yes, a preview can't be used.  But when you drag a preview to the workspace, it becomes a separate image which can be operated on just like any other image.  So starmask will work on the separate image just fine.

I've played with this some more and it has something to do with the noise threshold setting.  It's fine when working with the image created from the preview but not when working with the original image.  Setting the shadows to the value I was using for noise and changing the noise to 0.2 results in the starmask doing the expected thing on both the image created from the preview and the original preview...although not the same results as the original process settings applied to the image created from the preview.

So it's something about the two images that the same noise settings does different things in starmask.  I would have thought it would be the same behavior in both cases since the image created from the preview is an exact copy of a piece of the original image.  The noise setting, as I understand it, just says to ignore data below the setting when looking for stars.

Paul

[lucchett]

Paul, what I meant was that you can use previews but, as in your case, you get different results for the same settings.
the star mask launched on the original image and on the "preview cut" gives different results as the two images contain different pixels (in the preview there is a subset).
all the pixel in the images are used by the algorithm internal calculations, and for that reason the results are different.

Andrea

[lucchett]

Paul,
that was the Juan's post I was referring to.
Hope it's useful
andrea

http://pixinsight.com/forum/index.php?topic=665.msg2993#msg2993

[msmythers]

Paul

You might want to change the mode to Structure Detection. I find this works well with images like yours. Here is an example. The top half repeats your problem. I used a different noise threshold due to my image. The bottom half I use the Structure detection without changing the threshold. Now the stars in the center of the nebula are picked up. I then use a slight bit on convolution to round the stars in the mask since no smoothness is available for a structure mask.

Also remember that for difficult images you might need to make multiple masks and combine them in PixelMath for a more complete mask.

Mike

[SeaLandSkyPhoto]

Thanx for the info.  I'll give the structure detection mode a try.  'Twould be nice if previews worked reliably here since on my system its 30s for a starmask of the full image vs 1s for a starmask with the image from a preview. ;-)

Does anyone know where the post mentioned by Juan (following the link above, it has another link (talking about HDRMT with StarMask) http://pixinsight.com/forum/viewtopic.php?p=2830#2830 that leads to a not found page?


Paul

[lucchett]

Paul, I can't check right now but I remember the method has been discussed several times in the forum, using the search function should work.
I think it is also presented by Rogelio on his website (or a similar one).

basically:
-delinearize your luminance
-apply a strong HDRWT, may be two passes, to flat aggressively the image
-invoke the starmask tool, even with the default parameters, it gets all the small stars.

this works for the small medium stars that are usually an issue with the starmask tool.
you can then produce a second starmask for the largest stars using the SM tool or working with wavelets.
last you mix the two mask in Pixelmath with the max(a,b) function.

Andrea

[SeaLandSkyPhoto]

Thanx