Is my understanding that by 2x drizzle you reduce SNR by a factor of 2 because you only have 1/4th of the signal per "drizzle pixel" on average correct or am I missing something?
Thank you for starting this discussion. Highly interesting for me since I'm currently still creating very wide field work with camera lenses which is significantly undersampled. Will certainly give drizzling a try.
Clear skies
Wolfgang
If we are exposing our subs correctly, our noise is dominated by the sky noise, which is the shot noise due to the sky background (square root of sky background signal). With drizzle we split all signals over more pixels. In each pixel both the target and sky signal are affected equally. Updated; However, the image signal is reduced by 4 while the sky noise is only reduced by square root of 4, so there is a S/N loss of 2X. Also, there is some additional correlation noise between the finer pixels, which increases larger scale noise somewhat more: Fruchter and Hook As I mentioned in a previous post, the finer noise grain is easier to deal with and is not as objectionable as the original larger scale noise. So visually I do not see much of an increase in noise. Again this goes to making sure we get rid of pattern noise with dithering. And as well make sure our read noise is much less than our sky background noise. But these are things we should be doing anyway. It is more complicated with pattern and read noise, as those too are divided between pixels, but not so clear how those will be affected. But one should target minimizing these noises, so if we have your sub exposure tuned in well, these noises will not result in a significant increase in noise with drizzle.
Rick
|
You cannot like this item. Reason: "ANONYMOUS".
You cannot remove your like from this item.
Editing a post is only allowed within 24 hours after creating it.
You cannot Like this post because the topic is closed.
Rick Veregin:
But one should target minimizing these noises, so if we have your sub exposure tuned in well, these noises will not result in a significant increase in noise with drizzle.
I tried drizzling tonight for the first time and saw no noise increase due to it, but did see an improvement in detail and smoothness. I plan to use it in the future; not all the time, but when I really want to enhance a special image. It's not exactly hard from what I can see, but is a bit time consuming.
|
You cannot like this item. Reason: "ANONYMOUS".
You cannot remove your like from this item.
Editing a post is only allowed within 24 hours after creating it.
You cannot Like this post because the topic is closed.
Rick Veregin:
If we are exposing our subs correctly, our noise is dominated by the sky noise, which is the shot noise due to the sky background (square root of sky background signal). With drizzle we split all signals over more pixels. In each pixel both the target and sky signal are affected equally, so the S/N is the same.
The object to sky ratio does not change, but the SNR does decreases. Let's assume a uniform scene. If the sampling is increased and droplets are smaller than the original pixels, each pixel of the final image receives fewer photons than in the non-drizzled case. For example, with drizzle 2x2 and square droplets half the width of the original pixels, it takes a minimum of four dithered subframes to fill in the gaps. Thus, for a given number of subframes, each resampled pixel receives on average four times fewer photons than in the non-drizzled case, and the SNR in each is x2 lower. In addition, the noise between adjacent pixels becomes correlated, as discussed in section 7 of the Fruchter & Hook paper. How much of an increase depends on the ratio between resampled pixels and droplet sizes, i.e. how much the droplets fill the plane.
CS
Frédéric
|
You cannot like this item. Reason: "ANONYMOUS".
You cannot remove your like from this item.
Editing a post is only allowed within 24 hours after creating it.
You cannot Like this post because the topic is closed.
Frédéric Auchère:
Rick Veregin:
If we are exposing our subs correctly, our noise is dominated by the sky noise, which is the shot noise due to the sky background (square root of sky background signal). With drizzle we split all signals over more pixels. In each pixel both the target and sky signal are affected equally, so the S/N is the same.
The object to sky ratio does not change, but the SNR does decreases. Let's assume a uniform scene. If the sampling is increased and droplets are smaller than the original pixels, each pixel of the final image receives fewer photons than in the non-drizzled case. For example, with drizzle 2x2 and square droplets half the width of the original pixels, it takes a minimum of four dithered subframes to fill in the gaps. Thus, for a given number of subframes, each resampled pixel receives on average four times fewer photons than in the non-drizzled case, and the SNR in each is x2 lower. In addition, the noise between adjacent pixels becomes correlated, as discussed in section 7 of the Fruchter & Hook paper. How much of an increase depends on the ratio between resampled pixels and droplet sizes, i.e. how much the droplets fill the plane.
CS
Frédéric
... and we will not see more noise if we look at the images at the same size as undrizzled but should see more noise when comparing on pixel level.
Your explanation fits to what I was thinking but I just couldn't express it as precise and was not sure whether I'm missing something.
Clear skies
Wolfgang
|
You cannot like this item. Reason: "ANONYMOUS".
You cannot remove your like from this item.
Editing a post is only allowed within 24 hours after creating it.
You cannot Like this post because the topic is closed.
Frédéric Auchère:
Rick Veregin:
If we are exposing our subs correctly, our noise is dominated by the sky noise, which is the shot noise due to the sky background (square root of sky background signal). With drizzle we split all signals over more pixels. In each pixel both the target and sky signal are affected equally, so the S/N is the same.
The object to sky ratio does not change, but the SNR does decreases. Let's assume a uniform scene. If the sampling is increased and droplets are smaller than the original pixels, each pixel of the final image receives fewer photons than in the non-drizzled case. For example, with drizzle 2x2 and square droplets half the width of the original pixels, it takes a minimum of four dithered subframes to fill in the gaps. Thus, for a given number of subframes, each resampled pixel receives on average four times fewer photons than in the non-drizzled case, and the SNR in each is x2 lower. In addition, the noise between adjacent pixels becomes correlated, as discussed in section 7 of the Fruchter & Hook paper. How much of an increase depends on the ratio between resampled pixels and droplet sizes, i.e. how much the droplets fill the plane.
CS
Frédéric
Thanks very much Frédéric for the correction, I mispoke. Much appreciated! I'm updating my original comment. There is indeed a factor of 2x in SN as the sky background noise is the square root of the signal, so we have 4x less image photons/pixel, and 2x less sky noise, so an SN loss of 2X. Also, the noise is finer grain, so visually I typically do not see much of visible increase in noise at a reasonable viewing scale (zooming to the pixel level is pointless, as who cares about the noise when you can see the pixelation). Also, as mentioned in this thread, I often bin the image back in processing, this still maintains the FWHM advantage, better smoother stars and transitions, but improves the S/N.
Regards
Rick
|
You cannot like this item. Reason: "ANONYMOUS".
You cannot remove your like from this item.
Editing a post is only allowed within 24 hours after creating it.
You cannot Like this post because the topic is closed.
Andy Wray:
Rick Veregin:
But one should target minimizing these noises, so if we have your sub exposure tuned in well, these noises will not result in a significant increase in noise with drizzle.
I tried drizzling tonight for the first time and saw no noise increase due to it, but did see an improvement in detail and smoothness. I plan to use it in the future; not all the time, but when I really want to enhance a special image. It's not exactly hard from what I can see, but is a bit time consuming.
Hi @Andy Wray - could you provide details of your settings please? (Was this in APP?)
|
You cannot like this item. Reason: "ANONYMOUS".
You cannot remove your like from this item.
Editing a post is only allowed within 24 hours after creating it.
You cannot Like this post because the topic is closed.
Ian Dixon:
Andy Wray:
Rick Veregin:
But one should target minimizing these noises, so if we have your sub exposure tuned in well, these noises will not result in a significant increase in noise with drizzle.
I tried drizzling tonight for the first time and saw no noise increase due to it, but did see an improvement in detail and smoothness. I plan to use it in the future; not all the time, but when I really want to enhance a special image. It's not exactly hard from what I can see, but is a bit time consuming.
Hi @Andy Wray - could you provide details of your settings please? (Was this in APP?)
Sorry, but I use PixInsight and just used the standard settings. I did use the WBPP script in Pixinsight to create the drizzle files.
|
You cannot like this item. Reason: "ANONYMOUS".
You cannot remove your like from this item.
Editing a post is only allowed within 24 hours after creating it.
You cannot Like this post because the topic is closed.
By smoothing the transitions from objects to background, and being done before any post-processing, does drizzling impact subsequent processes like noise reduction, deconvolution, sharpening, star reduction, etc.?
Cheers,
Scott
|
You cannot like this item. Reason: "ANONYMOUS".
You cannot remove your like from this item.
Editing a post is only allowed within 24 hours after creating it.
You cannot Like this post because the topic is closed.
Scott Badger:
By smoothing the transitions from objects to background, and being done before any post-processing, does drizzling impact subsequent processes like noise reduction, deconvolution, sharpening, star reduction, etc.?
One side effect is that the noise in adjacent pixels becomes somewhat correlated. From my own experience, this generally has marginal impact. Processes like regularize RL deconvolution, which requires an estimate of the noise in the image, still perform satisfactorily without taking into account that correlation. In the example below (already posted, likely not the best, but it's mine  ), the subs were drizzled x4 with 0.5 droplet size, and regularized RL deconvolution was applied to get matching star sizes in the three color channels (see rev. B):
In short, I would say that you can process your drizzled images normally.
CS,
Frédéric
|
You cannot like this item. Reason: "ANONYMOUS".
You cannot remove your like from this item.
Editing a post is only allowed within 24 hours after creating it.
You cannot Like this post because the topic is closed.
6.65
10.98
3.61
7.75
6.10
7.61
