Author Topic: Problem in ABE  (Read 8157 times)

Offline LD

  • PixInsight Enthusiast
  • **
  • Posts: 86
Problem in ABE
« on: 2006 November 30 08:36:04 »
I don't know if this is a bug or user error--I have consistently gotten this same problem in both DBE and, now that I've tried it, ABE. (This is actually a follow-up to my previous thread on the general message section.)

I always seem to get no data at all in the upper right hand corner, as if corner of a page had been turned down to reveal a solid page underneath. I used the image of the edge-on galaxy to try out ABE because of it's very pronounced gradient. I was very, very amazed at how well using the default settings flattened out the image overall. But as you'll see, that upper right corner seems odd.

You can see the files here:

http://www.freewebs.com/drivewayastronomy/dbeimagepixinsight.htm

Thanks for any advice.

Larry Durst

PS: Any clues on how to upload files here--that way I won't have to put them on the website.

Offline Juan Conejero

  • PTeam Member
  • PixInsight Jedi Grand Master
  • ********
  • Posts: 7111
    • http://pixinsight.com/
Problem in ABE
« Reply #1 on: 2006 November 30 10:08:00 »
Hi Larry,

Sorry for the long delay in our answer.

As for inserting images in your posts here, I advise you to use imageshack (http://www.imageshack.us/). It is easy and works very well. You upload your image and they give you an url that you can use here, when composing your message.

The problem with your image is that it has zero pixel values. The generated background models (both DBE and ABE) also have zeroes, as one might expect.

If just isolated pixels had zero values in your image, then there is a high probability that the interpolated models would have no zeros at all. However, your image has an extended area with constant zero pixels at the upper right corner, which is necessarily reproduced by background models.

This is the explanation of what happens with PixelMath in PixInsight LE:

When you divide your image by one of those models, some divisions by zero occur. PixelMath assigns an "infinite" value (actually, the highest possible value in the range of IEEE 32-bit floating point) to those pixels that get divided by zero.

Finally, when PixelMath rescales the resulting quotient image, the "infinite" pixels cause a total lack of accuracy.

Solution:
Add a very small number to your image before generating the background model. For example, use PixelMath to add something like 0.001 (create a new row in PixelMath, select the ADD operator, set R=G=B=0.001, and select no image as operand; then disable the Rescale option). In this way you'll have no zero pixels in your image, and the extra 0.001 won't harm it.

Another option is subtracting the model instead of dividing. This is what you should do if what you are trying to fix with background modelling is caused by additive phenomena, like light pollution. Adding the model should not have any problems with zero pixels.

With ABE, the corrected image is not as odd as is with PixelMath in PixInsight LE's DBE. This is because ABE is smarter than LE to divide images.

The same is true with the new PixelMath in the PixInsight Standard application, which I encourage you to use. By the way, Carlos has implemented ABE as a process in the PixInsight Standard platform. Using it within the new GUI is a delight.

Hope this helps. Please let me know.
Juan Conejero
PixInsight Development Team
http://pixinsight.com/

Offline LD

  • PixInsight Enthusiast
  • **
  • Posts: 86
Problem in ABE
« Reply #2 on: 2006 November 30 10:54:36 »
Juan,
Thank you for that reply. I must say that I'm still figuring out how to use Pixel Math in the standard edition download (one of the things I like about ABE as a concept is that even I can figure it out  :wink: )
I will certainly give what you say a go.
Meanwhile, I look forward to your first commercial release. Your histogram and curves tools have become my standards for all images, and there is truly no other place to turn for noise reduction! I can only imagine what is feasible when the light clicks for me on all your other routines!
Best of luck,
Larry Durst

Offline Juan Conejero

  • PTeam Member
  • PixInsight Jedi Grand Master
  • ********
  • Posts: 7111
    • http://pixinsight.com/
Problem in ABE
« Reply #3 on: 2006 November 30 11:12:38 »
Hi Larry,

Thanks!  :D

Quote from: "LD"
...I'm still figuring out how to use Pixel Math in the standard edition download


Ok, here is a quick guide. The new PixelMath accepts expressions in common algebraic notation, with a few additional tricks. For example, if your image is Image01 and your background model is Image01_background, then write this in the R/K expression line:

Code: [Select]
$target / Image01_background

and apply the process to Image01. The $target will be replaced by the actual identifier of the image you're applying to (Image01 instead of $target would have worked the same in this example). To subtract the model instead of dividing:

Code: [Select]
$target - Image01_background

and so on. You can use parentheses, constants, lots of functions... PixelMath is extremely powerful now. I promise a tutorial about this tool very soon ;-)

Quote from: "LD"
I can only imagine what is feasible when the light clicks for me on all your other routines!


Well, the problem here is that we lack a reasonable documentation. Authoring a good documentation, for the basic topics at least, will be our main task during the next weeks. When we have it, all will be much easier. I know that many of our tools may look intimidating at first glance, but trust me that they are easy, once explained ;-)

Best,
Juan Conejero
PixInsight Development Team
http://pixinsight.com/

Offline LD

  • PixInsight Enthusiast
  • **
  • Posts: 86
Problem in ABE
« Reply #4 on: 2006 December 01 07:56:08 »
Thanks, Juan. Ah, algebraic notation--those numbers and symbols I pretend I'm actually helping my children with each night :wink:
Regards,
Larry