Can you dither too agressively?

I never dither so it is hard to say other than why should you? You haven't got a Canon DSLR either…

I just realised… I’m probably dithering not aggressively enough? I don’t have walking pattern noise but a fair share of resididual noise even with dithering and matching darks. Likely tho my lightpollution and small aparture combined with an IMX183 sensor play towards that as well, as I have less noise in Narrowband data or with very bright objects, probably higher SNR to start with

andrea tasselli:

I just realised… I’m probably dithering not aggressively enough? I don’t have walking pattern noise but a fair share of resididual noise even with dithering and matching darks. Likely tho my lightpollution and small aparture combined with an IMX183 sensor play towards that as well, as I have less noise in Narrowband data or with very bright objects, probably higher SNR to start with

Dithering only avoids fixed pattern noise and it doesn't take much displacement (few pixels would suffice) to avoid that. It does nothing to the overall noise floor.

Okay, thanks for clarification - so I can leave it as it is, thats good to know because Fixed Pattern Noise isn’t an issue for me since I started dithering.

andrea tasselli:
I never dither so it is hard to say other than why should you? You haven't got a Canon DSLR either...

I started to dither as everyone was saying "dither or die", but I actually didn't see the point of it with my mono CMOS sensor so gave up.  Am I missing anything here?

I'm dithering with about 20 pixels on the main camera and have no issues. So 65 pixels sounds a lot and why not reduce it if it doesn't hurt. I find Andreas comment interesting, is it really only Canon cameras? I know that they are particularly "bad" with their tendency for banding but I always thought that hot pixels can cause walking noise too. I'm referring to this discussion: Walking Noise: What is it? - Experienced Deep Sky Imaging - Cloudy Nights.
 
Clear skies
Wolfgang

I have no issue with FPN in any of my cameras, including the Nikons but excluding the Canons. There are 2 main reasons why I wouldn't bother dithering:

1. Unless for some miracle you have perfect PA you always end up with drift (in both directions). You can take advantage of that to avoid actual dithering.

2. Appropriate cosmetic correction is the cure for all the residual FPN you might have.

As I said before, I don't do it and I never felt the need of doing it. Processing takes care of it all and I'd rather waste less precious time with some of my mounts (e.g., the GEM28 has 3 min settling time!).

Dear all,
Thank to you so much all for your thoughts!
Next clear nights I am going to test different # pixels to see were the minimum amount of pixels needs to be.
Collegae of the Eindhoven astroclub looked at the data and saw that the result in the main camera on the x-axis varies between 12-170 pixel movement and on the Y-axis between 128-860 pixels so that seems a bit overkill for getting the positive effect of dithering.
Clear skies everyone!
Eugène

andrea tasselli:

I'm dithering with about 20 pixels on the main camera and have no issues. So 65 pixels sounds a lot and why not reduce it if it doesn't hurt. I find Andreas comment interesting, is it really only Canon cameras? I know that they are particularly "bad" with their tendency for banding but I always thought that hot pixels can cause walking noise too. I'm referring to this discussion: Walking Noise: What is it? - Experienced Deep Sky Imaging - Cloudy Nights.
 
Clear skies
Wolfgang

I have no issue with FPN in any of my cameras, including the Nikons but excluding the Canons. There are 2 main reasons why I wouldn't bother dithering:

1. Unless for some miracle you have perfect PA you always end up with drift (in both directions). You can take advantage of that to avoid actual dithering.

2. Appropriate cosmetic correction is the cure for all the residual FPN you might have.

As I said before, I don't do it and I never felt the need of doing it. Processing takes care of it all and I'd rather waste less precious time with some of my mounts (e.g., the GEM28 has 3 min settling time!).

But you also are losing the ability to do CFA Drizzle and have to fall back to debayering. That's a significant disadvantage in terms of recovering high frequency detail. Part of the reason so many people think the gap between mono and OSC is as wide as they do is that they're comparing non-interpolated data to interpolated data.

Roman Pearah:
But you also are losing the ability to do CFA Drizzle and have to fall back to debayering. That's a significant disadvantage in terms of recovering high frequency detail. Part of the reason so many people think the gap between mono and OSC is as wide as they do is that they're comparing non-interpolated data to interpolated data.

Absolutely not. It would work anyway because interpolation (debayering) occurs beforehand, at least in PI. And because main point n.1, i.e., you always drift for all practical purposes (the sum of drift and periodic error). And the gap isn't going to go way because you dither (or not).

David Redwine:
Dithering is necessary, especially with CMOS cameras, but it reduces your effective FOV. 10 pixels is plenty, try 3 or 5 to see if that works

It is necessary when the sensor diplays significant FPN otherwise ain't worth the bother. Or when you have very few frames, less than 3 per channel.

Dithering is a way to use the stacking filter to statistically minimize signals and patterns that are fixed with respect to the sensor.  Dithering helps with FPN but the correct way to deal with FPN is with flat-correction--not dithering.  Remember that FPN is the result of PRNU (Photo Response Non-Uniform) in the sensor so it's a source of uncertainty across the sensor that increases with signal strength.  Depending on the level of responsivity variation between signals, FPN can have a significantly larger impact on perceived noise than photon-noise, but that also depends on the brightness of the object.  FPN is multiplicative to the signal so it is corrected exactly like vignetting or radiometric fall-off by dividing the signal by the flat field data.  In fact, FPN is one of the three big reasons that we use flats in the fist place!  I have two QHY600M cameras.  One has almost no FPN (that I can see) but the other has a LOT, which requires good flat correction.

Dithering can help to further blur the effects of FPN but it is also very useful for removing residual dark noise and RBI ghost images around stars (and yes, both of these effects can happen in CMOS devices).  If you assume that the variations in FPN (and everything else) are mostly random across the sensor, the minimum dither distance is determined by the autocorrelation distance of the pattern.  That's the distance between two points on the surface where the values are statistically uncorrelated.  Unfortunately, many (most?) of these patterns are not randomly distributed and many such patterns have spatial structure.  That's why it's best to start with flat correction to remove FPN and to use a relatively large dither distance to "average out" any residual spatial patterns that may "poke through" the calibration process (for whatever reason--but usually due to variations with time).  The only way that a large dither distance hurts is when it becomes too large of a fraction of the total field.  If you keep it less than 1%-3%  of the full field, you won't be losing any significant amount of your field coverage by dithering over a big distance.

John

andrea tasselli:

Roman Pearah:
But you also are losing the ability to do CFA Drizzle and have to fall back to debayering. That's a significant disadvantage in terms of recovering high frequency detail. Part of the reason so many people think the gap between mono and OSC is as wide as they do is that they're comparing non-interpolated data to interpolated data.

Absolutely not. It would work anyway because interpolation (debayering) occurs beforehand, at least in PI. And because main point n.1, i.e., you always drift for all practical purposes (the sum of drift and periodic error). And the gap isn't going to go way because you dither (or not).

But CFA Drizzle doesn't use the interpolated, debayered data. Debayering is done to get you through standard integration which also produces drizzle files. The drizzle process using the raw monochrome CFA data. As for drifting that's not a generalization I think one can make and then give out as advice about best practices for noise reduction. If I don't dither, for example, my frames are damn near in lock step. Everyone's mileage will vary. Even so, drifting doesn't get you the same result as random dithering since it's directional and is often why people end up with "walking noise".

Ok. I checked the documentation and you're right. It does require the original calibrated/corrected CFA. This said, it does not require a random variation in displacements between frames but a non integer one, which is easily achieved when the images are not significantly undersampled. Additionally, you'd need a rather large number of files to make it work properly, which isn't something that is always achieved. The flip side is while providing better results in terms of alaising artefacts it does reduce the overall SNR (no free lunch). Besides, the raison d'etre of dithering is for dealing with FPN and random noise when number of frames is small. In this natural drift seems to me a suitable alternative to forced dithering IF the sensor isn't prone to produce FPN artifacts (e.g. Canon DSLR). As I said many times before I never dither and never ever had this sort of issues so I can only advise to dither if, after proper reduction techniques are applied, the data still show FPN in the usual form of "walking noise" or similar FPN artefacts.