›

DBE: You are in CONTROL

Sedat Bilgebay
in PixInsight
This is a great tip Adam! I'm almost sure that no one else has touched this subject. Thanks

Comments

  • BlockHeadBlockHead
    Thank you!
  • Wade Hilmo
    I have a quick question on this video.

    Near the beginning of the video, you showed the background model, created with the default sample placement.  In that model, the galaxy was clearly visible as a very bright glow in the center.  I knew from that, that you'd get a dark center in the corrected image.

    I was hoping that you'd show the background model as you altered sample points to see how it affected the glow in the center.  When I place samples with DBE, I always check the model for any hint of my object.  If I can see it, I adjust the samples.

    My question, is:  Would it have been equally effective to use the background model this way, instead of checking the corrected image?   I am guessing that if you adjust the samples to eliminate the glow from the background, that you'd also fix the dark center in the corrected image.
  • BlockHeadBlockHead
    Give it a go and let me know:


    -the Blockhead
  • Wade Hilmo
    Thanks for the data.

    I loaded the image and configured DBE with a tolerance of 10 and 20 samples per row.  I then moved each of the samples away from the edge.  To ensure clean samples, I reduced the tolerance back to 1 and then blinked through each of the samples and moved any that showed structure in the box.

    I then ran DBE to get a look at the background model and the galaxy was obviously visible as a glow.  I then started deleting any samples that were on or near the galaxy and built a background model to see the results.

    Through the process, I built a background model periodically to see if there was any trace of the galaxy.  When I got to the point where there was no obvious trace of the galaxy, I ran it with subtraction as the correction.  This left me with a few dark spots.  I overlaid the results on top of the image with samples and blinked them.  In each case, I found a poorly placed sample with something bright in it.  Once I moved those samples, I got a result that I was pretty happy with.

    I have uploaded my DBE settings and the final background and corrected image to dbe_test.zip at the following link:


    Here is a much reduced view of the result.  It's not perfect, but I think that it's pretty good.  I could probably improve it by using your technique with a few fixed samples around the galaxy, but I don't know if that's even necessary.  By the time the final black point is set, I think that the background will be pretty good "as is", especially since I only spent about 10 or 15 minutes on it.

    Based on this exercise, I do feel like using the background model as a guide is effective for efficient placement of samples.

    And thanks for the suggestion of giving it a try.  I feel like this is a much better answer than if you'd just given me your off-the-top-of-your-head opinion.

    -Wade




    NGC1300_Luminance_DBE.jpg
    1024 x 1024 - 129K
  • BlockHeadBlockHead
    Hi Wade,

    Nicely done. I looked at your result by applying your DBE sample pattern to my image.
    I think your technique in general has merit. I had never really considered it- mostly because I think
    it happens naturally with small numbers of samples that are widely spaced/placed. I am not a big fan of
    putting a gazillion samples down. But in this case, the halo of light was pretty yucky- so I felt I needed to based on prior experience. And ended up with the model discontinuity near the galaxy.

    Your technique is something like a hybrid of fewer samples- in the sense that where the object is, you avoid with carefully placed samples that create a smoother gradient across the area of concern. I think your result is pretty good and I will need to experiment with it.

    When I compared to my result, and this is really subtle, there is a difference. By placing the samples closer to the galaxy, the local sky is flatter right up the discontinuity. Your method effectively spreads the "darker" area across a wider region- which does flatten things pretty good. My method will certainly result in a "greater" subtraction across the galaxy- but these adjustments are pretty darn small- so I am not too concerned and more interested in the visual result. 

    So, I think I will definitely mention your technique as another way to monitor the goodness of the background model. The slight negative, is this it still resides in the "art" of sample placement- and I always want to favor methods that do not require this "skill" if possible. Your image is on the left, my result is on the right. 

    -the Blockhead

    Capture4.JPG
    2822 x 1428 - 803K
  • Wade Hilmo
    Normally, I like to use as few samples as I can.  It's very unusual that I have more than a dozen manually placed samples in a single image.

    This was in interesting case for me because of the uncorrected, and non-symmetrical, halo of light.  I was surprised that your technique in the video addressed it as well as it did.  I was even more surprised that my steps also corrected it as well as it did.

    I did want to point out that I was not trying to improve on your example of using fixed values for some set of samples near the galaxy.  My assertion was that using the background model is a good way to validate the sample placement and effectiveness.  It just happened that I was able to correct it by just deleting samples on the galaxy and moving a few others.

    Regarding the difference between our results, I was wondering about how each method affected the galaxy itself.  Unfortunately, I'm not sure that we can know for sure whether the halo is ring-shaped or disc-shaped.  If it were "natural" illumination, then I would assume disc-shaped.  I think that you need a flat field error to introduce a ring (which may or may not be the case here).  I suspect that your method removes more light from the galaxy than mine, but since it's in a high-signal area, that's probably ok.  I also think that if the central brightness is a result of natural illumination through the optical system, the both of our methods are technically incorrect.  The actual background in that case would be a circular gradient that continues to get brighter in the center.  My background model does not do that, since it's mostly flat in the center.  I've not seen the resulting background model after your final sample placement, but I suspect that it's a bit brighter than the natural background would be.  So you may be over-subtracting from the galaxy and I'm probably under-subtracting from it.  Again, since we're dealing with very small brightness levels, and the galaxy is a high-signal area, that's probably ok.

    And finally, regarding the "art" of sample placement, I'm not sure that there is a way around that.  I was trying to push it more towards the "science" end of the scale by reducing the tolerance and cycling through each of the samples to look for structure, and then bumping it back up to 10 for creating of the model and correcting the image.  I also think that blinking the corrected image against the original with samples, was pretty effective at finding the few remaining samples that included bright pixels, like small stars.  I clearly missed a few when examining the samples.  When I was doing it, I got the sense that hitting the "next" button on the sample selection skipped over some, and I guess that actually happened.  This was the first time that I tried blinking the corrected and original images that way, and I am happy with how it worked out.

    This was a bit of a rambling response, so I hope that it makes some sense.

    -Wade
Sign In or Register to comment.