Morphology Transformation Erosion creates filaments between stars

[jeffweiss9]

I have a longstanding PI problem that perhaps someone can tell me how to avoid. In star size reduction using a starmask and MT erosion with iteration, I find that with a dense field of very small stars, like those in this field containing sh2-101 in Cygnus,

http://www.astrophotogallery.org/member-galleries/p8218-sh2-101-tulip-nebula.html

MT starts to create filaments connecting the stars.  They are just starting to form in the linked image and get stronger and stronger had more iterations of 0.16 amplitude MT been performed which is what I wanted to do to get these small stars further reduced.   It seems clear that, if my star mask is too large for these smaller stars, then the masks from adjacent stars touch or nearly touch, then the erosion operation will erode the outer portions of each star except on a line connecting adjacent stars. It is completely mysterious to me, however, how the erosion operator somehow ends up creating light in areas that were originally dark to build up a totally-connecting filament, even with an oversized mask.  Erosion should only erode. 
  Any help, suggestions, explanations appreciated.
Thanks.
-Jeff

[pfile]

hi, me again

another thought i had was that this is what MT can look like when it is applied with no mask. i know at least once i had applied my star mask, but accidentally disabled it. the only indication of this is "Masked (Disabled)" down in the status bar. you'd think that removing a mask and applying it would reset the enable/disable state of the masking, but it is sticky.

anyway assuming this is not the problem, it might be helpful to post a screenshot of the image with the mask applied, just to see how open/closed the mask itself is.

[jeffweiss9]

Here are some more detailed pix, but the masks were definitely applied and this is a 'standard problem' of mine; for different masks, it's only when, rather than if, the filaments start to show up with MT iterations (typically 0.16 amplitude per iteration).
Pix's:
1) Picture before next MT   (what I'm now posting as just prior to filaments)
2) Picture with MaskApplied (picture with mask visible)
3) Picture with MaskApplied closeup
4) Picture after next MT (actually 6 more iterations) showing the dreaded filaments
This effect has kept me from getting the star reduction I want to sufficiently suppress these myriad small stars.
Thanks for your help, advice.
-Jeff

[oldwexi]

Hi Jeff!
Had similar results with morphology and erosion - until Rogelio posted this:

http://blog.deepskycolors.com/archivo/2011/09/08/star-size-reduction-via-Morphological-.html

Thats the perfect star reduction path. No artefacts. Does not use Erosion!

Aloha
Gerald

[pfile]

well, unfortunately i think that's where jeff started off, with RBA's tutorial. do you have the original image available someplace? maybe we can play with it.

[jeffweiss9]

Yes, I've already been talking to Rogelio about this image that attempted to use his tutorial as guide for star reduction (I'm very lucky to have him as a friend who happens to also live in the same town).  He has seen the same filamentary phenomenon in early images and suggested in this case using pixel math to subtract the starmask rather than use MT erosion.  I did that, empirically finding the right coefficient that removed/reduced small stars the most before introducing dark specs into the image.  This suggestion was successful in getting more reduction than I could with MT due to its introduction of the dreaded filaments so this definitely represents a viable alternative.   After doing some more work to enhance the tulip over the remaining small stars, I've posted the reprocessed version at:

http://www.astrophotogallery.org/member-galleries/p8218-sh2-101-tulip-nebula.html

I'm a lot happier with that one, although I don't know if the expert eyes here will agree.  The problem that I saw fundamentally was to pull out that tulip buried in such a dense starfield, and I think this is a lot closer to achieving that goal than I was before.

Looking more closely at my data, I did see some tendancy toward 'proto-filaments' already there before MT, so I am thinking that I need to do some deconvolution in my PI workflow.  That is something I have never gotten to work in PI, so that looks like one new thing to work on.  Perhaps this is the answer to my question of how the PI MT Erosion can end up creating light areas between closely-spaced stars where there were none to start with.  Or perhaps Juan or one of the PI authors can explain how MT is doing that, so perhaps we can prevent it from happening.

I can't post the pre-star reduction image here for size limitations but will 'yousendit' to Rob and Rogelio, or anyone else that asks for it.

Thanks very much for your help and advice.

-Jeff

[Juan Conejero]

Hi Jeff,

MT starts to create filaments connecting the stars.  They are just starting to form in the linked image and get stronger and stronger had more iterations of 0.16 amplitude MT been performed which is what I wanted to do to get these small stars further reduced.   It seems clear that, if my star mask is too large for these smaller stars, then the masks from adjacent stars touch or nearly touch, then the erosion operation will erode the outer portions of each star except on a line connecting adjacent stars. It is completely mysterious to me, however, how the erosion operator somehow ends up creating light in areas that were originally dark to build up a totally-connecting filament, even with an oversized mask.  Erosion should only erode.

And it does. However, bear in mind that when you apply several iterations of MT with amplitude < 1, you aren't applying just a morphological operator, but a recursive procedure that mixes processed and unprocessed data at each iteration. So it is not just erosion.

[jeffweiss9]

Hi, Juan-
  Ok, I had earlier convinced myself somehow iterating with small amounts in MT was better for star reduction but I'll guess I'll rethink that one and try again.  I needed some noise reduction in that image and think background noise fluctuations might have been 'seeding' the growing filaments.  Meanwhile, I did 'complete' that sh2-101 image with ACDNR and using a Pixel Math subtraction, at Rogelio's suggestion, instead of the MT erosion, which did get me further star reduction in this case.  My result (humbling to post here) is at:

http://www.astrophotogallery.org/member-galleries/p8218-sh2-101-tulip-nebula.html

-Jeff