variable sky gradient and drizzle

[whwang]

Hi,

I have a question that kind of bothers me.

I want to use drizzle to combine images, either Bayer drizzle, or standard drizzle, or even both.  This will require many exposures to fill in the pixel holes in the drizzle plane.

Now, suppose I can't accumulate enough exposures in a single short session.  I either will have to collect the exposures in a very long session, or across different nights.  In such a case, it is very easy to imagine that the sky gradient in the images will change from exposure to exposure, as the target moves on the sky.  Some of the exposures may be brighter in the south side, while others may be brighter in the north side.  And the brightness an the strength of gradient will also change.

I know PI can normalize the brightness of all the exposures and stack them.  How about the gradient?  Can drizzle still work well when the images have different gradients?  Even if overall the images of equal brightness after PI's normalization, the corner brightness will be different from exposure to exposure because of the variable gradient.  In such a region where the brightness are essentially not normalized, can drizzle still work well?  If not, what would you suggest to overcome this?

Cheers,
Wei-Hao

 

[whwang]

Now I am quite sure I am haunted by the variable sky gradient.

Is there anyway to do a gradient removal first before images are Bayer-drizzle combined?  For normal drizzle, I think this is quite straightforward.  I can just run ABE on each of the registered images and then run the drizzle integration.  For Bayer-drizzle, how can this be done?  Same as the normal drizzle?

 

[whwang]

Looks like I was wrong a little bit.  For the normal drizzle, I will have to go back to the calibrated/light/debayered directory to conduct a gradient removal, not the registered directory.  This must be still manageable, as the debayered images can be easily gradient-removed with ABE or DBE.

On the other hand, for Bayer drizzle, the images to be used are those in the calibrated/light/bayered directory.  There, the images can't be fed to ABE.  Anyway to solve this?

ALOHA, anyone in the house?

 

[Juan Conejero]

Hi Wei-Hao,

ALOHA, anyone in the house?

Yep, Monday 08:58 am here

Can drizzle still work well when the images have different gradients?

Drizzle just applies image weighting and normalization parameters computed by ImageIntegration. So the gradients can only be a problem for normalization during integration, as happens with regular interpolated, non-drizzled integration tasks.

The linear fit clipping pixel rejection algorithm is very robust to gradients with differing orientations and intensities. So unless your light pollution gradients are extreme (and in such case you have much worse problems with the data set), linear fit clipping will work remarkably well for rejection. You need a sufficiently large set of images for linear fit clipping to work properly, though.

As for image normalization, again if your gradients are reasonable there should be little problems, since ImageIntegration implements robust location and scale estimators. In general, the achieved integrated result will be nearly optimal if you apply ImageIntegration correctly to optimize rejection and maximize effective noise reduction.

Finally, as you have discovered, you can apply background modeling and correction to the debayered frames as an option. In practice, however, this is normally not necessary unless you are dealing with extreme and strongly varying light pollution gradients. Bayer drizzle will not be different at all in this regard, since you already have to work with debayered and registered frames to perform the image integration task before drizzle.

 

[whwang]

Hi Juan,

Yes, the gradient is variable in the particular case I am working on.  It doesn't happen often, but it did happen this time (low elevation plus light pollution plus being close to twilight).  The gradient is bad, but the data are still useful.  I don't want to throw away data. 

In such as case, the global image normalization cannot match the background brightness in certain corners of the image.  In other words, locally, the images are not normalized.  This produces problems for both normal drizzle and Bayer drizzle.  In the normal drizzle case, I can now make the problem go away by running ABE on each images in the debayered images.  In the Bayer drizzle case, this does not work, as ABE failed on the bayer images.  All these have nothing to do with pixel rejection.

If my way of thinking is correct, this problem can be solved if ABE can recognize the data in the bayer images and run gradient removal correctly.  Any chance this will happen in the future versions?

Cheers,
Wei-Hao

 

[whwang]

Finally, as you have discovered, you can apply background modeling and correction to the debayered frames as an option. In practice, however, this is normally not necessary unless you are dealing with extreme and strongly varying light pollution gradients. Bayer drizzle will not be different at all in this regard, since you already have to work with debayered and registered frames to perform the image integration task before drizzle.

Hi Juan,

I read the above again and tried to understand.  But then I realized what you said does not match what happened in my case.

There are three sets of images I potentially need to deal with: images in the directories of:
1. calibrated/light/bayer/
2. calibrated/light/debayered
3. registered

I first ran ABE on all images in #3.  (And normal image integration for pixel rejection with the drizzle files added.)  Then I carried out both Bayer drizzle integration and normal drizzle integration.  I found the drizzle-integrated images I got still have strong gradients.  The ABE did not seem to work.  Then I realized that the normal drizzle actually uses images stored in #2 and the Bayer drizzle uses images stored in #1, not in #3 in either case.  This is why I said my post on Jan 6 had some mistake.

Once I ran ABE on all images in #2, the normally drizzled images does not have a gradient in the background any more.  The trouble caused by variable gradient (grid pattern in the final image) also went away.

So I tried to do the same to Bayer drizzle, i.e., ran ABE on all images in #1.  And then I hit a wall.  ABE refused to run on images in directory #1.

So in short, Bayer drizzle does not seem to be dealing with the registered and debayered images.  It deals with the unregistered bayer images.  This is different from what you said.

Cheers,
Wei-Hao