I recently watched Richard Bennion's CCDStack tutorial. In the tutorial he went through processing steps to produce an image. I have a question about one of the steps. Following registration he did a resample step. If I understood the process the resample was at the same scale, in other words, it did not change the scale of the image. What is the purpose of this resample step? Is it required?
George

quote:

... Following registration he did a resample step...

That is actually the final stage of "registration/alignment" and it occurs in the "Apply" step.

In order for an image to be aligned to another image, it must be resampled (i.e. each pixel in the aligned image must come from one or more pixels in the original image). CCDStack gives you a choice of reampling methods. (All other CCD Software must also resample, but many of them do not make it explicit).

Stan

When you take a stack of images and match them to a reference image you are doing two things:

1. Registration

This is where the program calculates the difference of the scale, rotation and shift of each image in the stack as compared to the reference image.

This can be done three ways:

A. Auto Shirt - CCDStack will automatically calculate the registration difference, but will only shift the image (no rotation or scale).

B. Star Snap - This is where you select stars in the reference image and CCDStack automatically determines the registration difference of all images in the stack. This includes shift, scale and rotate offsets.

C. Manual - This is where you can manual shift, rotate and magnify (scale) the image yourself. This is a very good way to match up images at different focal lengths.

Helpful hint: You can roughly position an image manually using (c) above, then switch over to Star Snap (b) to set reference stars and have CCDStack do the precision alignment for you. I use this technique for matching up different focal length exposures.

Once all the images are registered (which is much easier executed then described) all of the offsets are now stored in computer memory for each image in the stack.

But, the images have not been permanently altered at this point, just the offsets are calculated. So we need now to make physical changes to each and every image in the stack to match the reference image (which is not effected by the next step).

2. Resampling:

Now that we know the offsets for each image in the stack to make it register with the reference image, we are now ready to alter these images physically so they are ready for data rejection and stacking.

You do this by clicking on the Apply tab in the Registration dialog box. You then have a choice on how you want to resample your images.

The help file provides information on each one of the three options, but let me add to this a bit based on my experience. Stan may have some more insight on this as well.

Scenario # 1:

If your images are properly sampled i.e. less than one arc-second image scale, and you are very well polar aligned, you should have images that require very little rotation or scale offsets.

In this case, use Auto Shift + Nearest Neighbor resampling. This will preserve all the detail in each image. Combine this with a Poisson data rejection procedure and you will preserve your stellar profiles and detail as best as possible.

Scenario #2:

You are working at an image scale of greater than > 1 arc-second and/or you are not very well polar aligned.

Use Star Snap + Bicubic b-spline. This keeps the star tight but will provide a smoothing effect on all the resampled images as we are now making adjustments for scale, rotation and shift using sub-pixel accuracy.

If you have a lot of images i.e. greater than 5, use a sigma data rejection. If you have greater than 10 images, use a min/max clip.

This will increase stellar profiles slightly and will provide a smoothing effect on the image resulting in a small (5%) loss of detail.

Method #3:

Your images are undersampled i.e. 2 arc-seconds or higher. In this case, you may want to use Auto Shift (if polar aligned well) or Star Snap (if polar alignment suffers or you are combing images over multiple nights) and bilinear resampling as this seems to keep the stellar profiles smaller in undersampled images and provides the cleanest noise model.

Now all this is from my experience with CCDStack.

Stan, please feel free to jump in here and correct me if I am mistaken or if there are better ways to achieve better results.

Best,

Richard, Stan,

Thanks for you detailed and prompt reply. I have been using CCDSoft or AIP4Win for alignment prior to combining with Sigma. I assume from your comments, the resample comes automatically with these applications when I do an alignment. Not sure what kind of reaampling is used by either of these two applications. I have used CCDInspector to examine the aligned images. The FWHM seems to increase with the alignment process, moreso for AIP4win than for CCDSoft. Both seem to increase the FWHM 5-10%. Is this normal? can this be avoided? Does CCDStack also produce this effect? (my trial has expired).

George

quote:

I assume from your comments, the resample comes automatically with these applications when I do an alignment.

Correct, this is a very simple automated process.

quote:

Does CCDStack also produce this effect?

Here is my experience. A stack of images that range from 1.9" to 2.6" in a fairly linear manner, will produce a final image with a FWHM of 2.3" using nearest neighbor method. With bicubic b-spline the FWHM will be around 2.5".

My comparitive data suggests CCDStack preserves FWHM better than any other registration routine on the market.

rb

quote:

...increase the FWHM 5-10%. Is this normal? can this be avoided? Does CCDStack also produce this effect?

That is unavoidable to some degree, though there are some "resampling" algorithms that also do an unsharp mask (which can be done is 2 steps in CCDStack). This is one good reason to avoid undersampling.

In CCDStack you should choose the method that is optimal for your data set. That's a slighly complicated subject but here are some recomendations:

1) if you are shifting only (no rotation or magnification) then bi-linear will probably give you the least blurring on each image.

2) if you are rotating or magnifying then bi-cubic will usually produce the best results for each image.

3) if there are at least 8 images in your stack and they are not undersampled then you might get best final results from "nearest neighbor". Each individual image may look funky but they may combine into a good image.

quote:

(my trial has expired)

If you download the latest version then you can still use CCDStack past expiration, except that all file saves are disabled.

Stan