DBE question

[greeko]

Hi

When using the DBE tool to generate the sampling of an image the result is lacking samples over the entire image and I'm not sure how to compensate for this.
On occasion I've had the entire image covered but not sure what feature function or parameter has an impact on this.

The most common setting I use are
Default sample radius 12
Samples per row 20

I"ve tried different settings with limited change.
Is there some parameter or pre cursor to help ensure the samples are generated to cover the entire image?

The idea is to eliminate having to manually generate the samples which isnt fun on a APS size chipset image.

Thanks in advance.

Troy

[Carlos Milovic]

Hi Troy

If you want something automatic, and just play with some parameters, give first ABE a try. It works with another kind of interpolation, but it makes a global fit, instead of a local one, as in DBE.

Of course, ABE will fail if you images are not dominated by background sky, i.e. you have large nebulaes, or very dim objects crossing the entire frame. In those cases the only choise is DBE.

Now, in DBE, you have 2 choises. Using a system similar to what is internally done in ABE, DBE lets you automatically generate samples over the entire image. To include (or reject) more samples, just change the tolerance and shadows relaxation parameters. How dense is the grid is controlled by the number of samples per row. Usually it is not necessary to cover the whole image, or to have too many samples. Just enough to be representative. Remember that the background modelization is a smooth function, and in normal conditions you wont have small or medium scale variations.
Once you found a good compromise with those parameters, manually add or erase samples, to ensure that they are located over background sky only. This is the main advantage of DBE over other algorithms.

[David Serrano]

manually add or erase samples, to ensure that they are located over background sky only.

Is this really needed? I understand that the box in the DBE's interface where the samples' contents is shown is lighter in areas that have more weight, and darker in areas that tend to be ignored for the purpose of the model generation. So, a sample that falls on a star usually has a black circle, indicating that there's no problem at all with it being on a star. This also implies that samples over nebulosity have less weight too.

Am I wrong?

[Carlos Milovic]

Yes, stars don't count. Very large ones are automatically excluded, so it is very unusual to have a sample over one of them (the internal algorithm rejects the sample). I was talking about really faint features, wich are very close in brightness to the background, so it is not obvious for the algorithm to discriminate, even with a low tolerance. Also this scheme may be screwed a bit if the object is too big (cover a significant portion of the image), so the global statistics are not a good starting point to determinate what is considered as background.

BTW, less weight does not mean no weight at all It may produce a slight deformation on the output model, even when the current DBE's algorithm is much more robust and intelligent right now to avoid this side-effect.

(Have I said that sometimes I'm quite a bit "perfectionist"? )

[Harry page]

Hi Troy

When I get this problem I adjust the min sample weight down to get the samples on the screen and once I have done
that fiddle with the tolerance setting to get a good selection in samples ( Ie lightish in colour)
Most stars are ignored but i would steer away from bright stars


Regards Harry

[greeko]

Hi Carlos and Harry

This was really helpful and thanks for your suggestions I'm rolling again.
No carpeltunnel for me now.

Troy

[Nocturnal]

It's not the cores of the stars that are the problem but the faint outer regions. Depending on the image and the DBE settings these can affect the model. I avoid stars in my samples and only place them by hand. If you get areas that are lighter or darker than the surroundings look for samples in that are in the wrong spot. If a sample is too light the resulting model will be light which in turn punches a 'hole' in the image.

With the new DBE (integrated image adjustment) it's much easier to make iterative improvements on the DBE data set.

[Juan Conejero]

I fully agree with Sander's opinion: outer regions of stars are the most problematic ones for DBE's rejection routines. The reason is simple: uncertainty. Statistical models for real (that is, uncertain) data are always subject to error (that's why statistics isn't actually a branch of mathematics ). So we really can't discriminate where a star ends and the background begins in an "ideal" way.

The Tolerance DBE parameter, along with a few more, has been designed to deal with this problem. Tuning Tolerance is actually a must to achieve good results with DBE. The same is true for ABE, of course.

[Nocturnal]

Hi,

I should add that high quality optics, mount and clear skies all make this less of a problem because the FWHM values of star go down. This makes them steeper and that allows statistics to be more effective as that faint outer region gets smaller and smaller. So now you know why I need to be careful with stars in my samples So-so optics, crappy mount and terrible skies