| Full frame image, TOA 150 + flattener, STL11K camera 55 min L |
| Perseus A Region Resolution Tests |
 |
 |
This project started out as a comparison of two OTAs of differing design and aperture (Tak TOA 150 and Vixen VC200L).
It quickly became apparent that the effects of registration and of resampling to comparable image scale caused at least as much difference in
image quality as any difference in the instruments. This was a revelation. I learned I had developed several ways to ruin good data through
registration and processing techniques!
Here is the full frame of the Perseus A region with the Tak TOA-150. Perseus A is a radio source corresponding to the marked galaxy, NGC
1275, a Seyfert Galaxy.
It quickly became apparent that the effects of registration and of resampling to comparable image scale caused at least as much difference in
image quality as any difference in the instruments. This was a revelation. I learned I had developed several ways to ruin good data through
registration and processing techniques!
Here is the full frame of the Perseus A region with the Tak TOA-150. Perseus A is a radio source corresponding to the marked galaxy, NGC
1275, a Seyfert Galaxy.
| Detail from above image with NGC 1275 marked |
This was a resolution test (hence the choice of a field with numerous stars and many small structures but not a long acquisition). As I started
processing, paying more attention to the numbers than usual, I first realized that resolution was highly dependent on Alt of image acquisition. I hadn't
realized before starting this project that FHWM is proportional to Air Mass.
processing, paying more attention to the numbers than usual, I first realized that resolution was highly dependent on Alt of image acquisition. I hadn't
realized before starting this project that FHWM is proportional to Air Mass.
Usually I shoot starting at an Alt of 30 degrees but note how even an alt of 40-45 degrees with an air mass of 1.4-1.5 bumps the FWHM up from the mid
2's to the mid 3's!
Think how hard you work on autoguiding, tracking, flexure and focus to get your FWHM from 2.9 down to 2.4. But take some of your subs at 45 degrees
of Alt and you will bump that FWHM up to 3.6". If the target is low in the south you are stuck, but if it passes near zenith you may be better off taking
fewer subs per night and limiting those to subs near the meridian.
2's to the mid 3's!
Think how hard you work on autoguiding, tracking, flexure and focus to get your FWHM from 2.9 down to 2.4. But take some of your subs at 45 degrees
of Alt and you will bump that FWHM up to 3.6". If the target is low in the south you are stuck, but if it passes near zenith you may be better off taking
fewer subs per night and limiting those to subs near the meridian.
1.69"/px. As I up or down scaled the data from one to match the other, resizing lost considerable resolution.
Looking into the docs for CCDStack I realized all this was spelled out (if I had ever bothered to read the docs). It then occurred to me that this effect of
resampling also applied to that resampling used to register images. Even though registration may not change the scale of the image, it will resample
pixels from the target image to match pixel locations in the master image. So I tested the various registration resampling methods to see what effect
they each had on FWHM.
Which resampling method is "best" isn't a simple choice. Some work well for images that are of the same scale (not mixing data from two different
scopes and not up/downsizing the image) and some work better with rotational matches. Others work best with pure translocations as in dithered
images.
The table below shows the FWHM of a single sub each registered to the same Master Frame by each of the various algorithms in CCDStack. The first
file listed is the original image before registration. Note it has a FWHM of 2.93".
NN= Nearest Neighbor, Lanc 256 or 64 = Lanczos/sinc, BL = Bilinear and Quad = Quadratic The rest are obvious:
Looking into the docs for CCDStack I realized all this was spelled out (if I had ever bothered to read the docs). It then occurred to me that this effect of
resampling also applied to that resampling used to register images. Even though registration may not change the scale of the image, it will resample
pixels from the target image to match pixel locations in the master image. So I tested the various registration resampling methods to see what effect
they each had on FWHM.
Which resampling method is "best" isn't a simple choice. Some work well for images that are of the same scale (not mixing data from two different
scopes and not up/downsizing the image) and some work better with rotational matches. Others work best with pure translocations as in dithered
images.
The table below shows the FWHM of a single sub each registered to the same Master Frame by each of the various algorithms in CCDStack. The first
file listed is the original image before registration. Note it has a FWHM of 2.93".
NN= Nearest Neighbor, Lanc 256 or 64 = Lanczos/sinc, BL = Bilinear and Quad = Quadratic The rest are obvious:
Unfortunately just minimizing the degradation of FWHM isn't the whole story. Some algorithms produced a small FWHM but created artifacts. And not
surprisingly the methods with the _worst_ FWHM produce the roundest, most perfect star shapes because they made large fuzzy balls out of a tight stars,
but these methods blurred detail in galaxies and nebulae at the same time.
Below is an animation of a few of the methods. Nearest Neighbor seems the best for this data (all at the same image scale, all from the same scope and
reducer and the only rotations 180 degrees due to meridian flip. Bicubic, which I had used up to this time, gives an obviously fuzziness to the stars and
loses detail in the galaxies. Bicubic 16 produces a reasonably small FWHM (3.06") but creates a dark pixel at the 6 o'clock position of bright stars.
That dark blob is a single pixel, by the way, showing how magnified these images are. Much of this would not be visible in a full frame image.
I have similar images of all the CCDStack registration methods but those are not posted here due to file size. The animated GIF is about 30 Mb, but I can
email a 3 Mb ZIP of the images to interested parties if you contact me.
surprisingly the methods with the _worst_ FWHM produce the roundest, most perfect star shapes because they made large fuzzy balls out of a tight stars,
but these methods blurred detail in galaxies and nebulae at the same time.
Below is an animation of a few of the methods. Nearest Neighbor seems the best for this data (all at the same image scale, all from the same scope and
reducer and the only rotations 180 degrees due to meridian flip. Bicubic, which I had used up to this time, gives an obviously fuzziness to the stars and
loses detail in the galaxies. Bicubic 16 produces a reasonably small FWHM (3.06") but creates a dark pixel at the 6 o'clock position of bright stars.
That dark blob is a single pixel, by the way, showing how magnified these images are. Much of this would not be visible in a full frame image.
I have similar images of all the CCDStack registration methods but those are not posted here due to file size. The animated GIF is about 30 Mb, but I can
email a 3 Mb ZIP of the images to interested parties if you contact me.
So what about the VC200L? That was the point of all this. The VC200L subs do not have quite the FWHM of the TOA subs. FWHM with the VC200L
(before any registration or resampling) was on the order of 2.9-3.5" compared to about 2.5" or so for the TOA. Before matching image scales the TOA
image _looked_ sharper because the galaxies, stars and nebulae were all minified.
Once I corrected for the image scale, which I now could do without inducing much loss of resolution, the TOA still has a sharper image when magnified to
the image scale of the VC200L:
(before any registration or resampling) was on the order of 2.9-3.5" compared to about 2.5" or so for the TOA. Before matching image scales the TOA
image _looked_ sharper because the galaxies, stars and nebulae were all minified.
Once I corrected for the image scale, which I now could do without inducing much loss of resolution, the TOA still has a sharper image when magnified to
the image scale of the VC200L:
54"). Note the VC has larger aperture (200mm vs 150mm), longer FL (1800mm vs 1100mm) and better sampling (1.03"/px vs 1.69"/px). But it
also has a large central obstruction and is a much less expensive scope implying (one would hope <G>) better attention to workmanship in
the TOA. Furthermore, the VC is subject to collimation errors.
I _think_ my VC200L is well collimated. I have asked in the VC group if anyone is getting FWHM in the mid 2's or better. I get ~3" FWHM with
the VC200L. As yet I haven't heard back that anyone has better results. I need to test this with an OTA with longer and better sampling. The
VC200L is not all that much different in FL or image scale than the TOA.
What are my conclusions:
1) Regarding the original question, this particular 6 inch refractor gives finer detail than a longer FL, larger aperture compound scope. You
have to upscale the image to get the same image scale but if you do that carefully you will still have better resolution with a zoomed image
from the 6" refractor.
2) The 6 inch refractor does have a disadvantage on dim targets. But with automated imaging it is pretty easy to get more hours. Nothing will
fix a blurry image.
3) The surprise important lesson was that it is really easy to lose resolution in registration. You fuss with autoguiding focusing and a dozen
other things to get your FWHM from 2.9" to 2.2' and then if you use the wrong registration algorithm while stacking your subs the FWHM
blows up to 3.6" in spite of all your other efforts.
4) Since you will lose at least some resolution with registration, be sure only to register your images once. Neil Fleming stresses that on his
website. (www dot flemingastrophotography dot.com). Don't register the subs to make a master red, master blue etc, then register those to
each other. Make a Master frame from the data from one filter, then use that master to register the subs from each of the other filters.
At every step from collection of the starlight to the finished image you have to fight to lose as little resolution per step as possible if you want a
decent outcome.
Drew Sullivan
New Years Day 2009
also has a large central obstruction and is a much less expensive scope implying (one would hope <G>) better attention to workmanship in
the TOA. Furthermore, the VC is subject to collimation errors.
I _think_ my VC200L is well collimated. I have asked in the VC group if anyone is getting FWHM in the mid 2's or better. I get ~3" FWHM with
the VC200L. As yet I haven't heard back that anyone has better results. I need to test this with an OTA with longer and better sampling. The
VC200L is not all that much different in FL or image scale than the TOA.
What are my conclusions:
1) Regarding the original question, this particular 6 inch refractor gives finer detail than a longer FL, larger aperture compound scope. You
have to upscale the image to get the same image scale but if you do that carefully you will still have better resolution with a zoomed image
from the 6" refractor.
2) The 6 inch refractor does have a disadvantage on dim targets. But with automated imaging it is pretty easy to get more hours. Nothing will
fix a blurry image.
3) The surprise important lesson was that it is really easy to lose resolution in registration. You fuss with autoguiding focusing and a dozen
other things to get your FWHM from 2.9" to 2.2' and then if you use the wrong registration algorithm while stacking your subs the FWHM
blows up to 3.6" in spite of all your other efforts.
4) Since you will lose at least some resolution with registration, be sure only to register your images once. Neil Fleming stresses that on his
website. (www dot flemingastrophotography dot.com). Don't register the subs to make a master red, master blue etc, then register those to
each other. Make a Master frame from the data from one filter, then use that master to register the subs from each of the other filters.
At every step from collection of the starlight to the finished image you have to fight to lose as little resolution per step as possible if you want a
decent outcome.
Drew Sullivan
New Years Day 2009
