M33 posted to help me learn.

[vlaiv]

10 minutes ago, Rustang said:

Your a clever man but most of the time your knowledge is wasted on me as it goes right over my head! 😁

At least there is simple point to all that - do dither it is good for the final image

[Rustang]

12 minutes ago, vlaiv said:

At least there is simple point to all that - do dither it is good for the final image

I've just got to work out if I've got the correct settings in APT and PHD2 now! 😊

[Rustang]

14 minutes ago, Aramcheck said:

On @alacant's suggestion of stacking without Darks, you can of course try this & compare the noise levels on a background area. I tried it once & found that with Darks included in the process, my background noise with my astromodified 600d appeared to be about half that compared with the same set of Lights stacked with only Master Bias & Master Flats.

BTW I also have the problem with Dithering timing out on occasions.

Cheers
Ivor

So to clarify on your test, you stacked a dithered set of images with and with out Darks and the noise appeared to be less when you still added Darks? 

[wimvb]

1 hour ago, vlaiv said:

Not true.

Dithering is very beneficial for lowering noise in the image, although most people don't know this. This is especially true when small number of calibration subs is taken.

To best explain this we need to look at what is happening when dithering. Let's observe single pixel and part of target on that pixel. If we have perfect tracing, same piece of target always lands on same pixel - with dithering it is always different pixel.

This means that with perfect tracking stack of pixels for our part of target come from single pixel in each sub and consequently get calibrated always with same bias value - bias is in this case constant and does not average out beyond it already being stacked and averaged.

With dithering - target covers always different pixel and bias sub calibrating always different pixel will have always different value as residual after stacking is random - this makes bias noise that we inject back in the image much smaller than in above example.

 

I guess that is what I meant with

1 hour ago, wimvb said:

But dithering will spread any fixed pattern noise,

 

[wimvb]

1 hour ago, vlaiv said:

This means that with perfect tracking stack of pixels for our part of target come from single pixel in each sub and consequently get calibrated always with same bias value - bias is in this case constant and does not average out beyond it already being stacked and averaged.

But, doesn't the light sub also contain this bias signal? If so, Bias(light) - bias(bias) =0

Or, bias(light) - bias(dark) = 0, and bias is canceled in calibration. And isn't that the whole point of calibration?

Edited October 12, 2020 by wimvb

[vlaiv]

3 minutes ago, wimvb said:

But, doesn't the light sub also contain this bias signal? If so, Bias(light) - bias(bias) =0

Or, bias(light) - bias(dark) = 0, and bias is canceled in calibration. And isn't that the whole point of calibration?

Signal yes but not the noise.

Calibration removes signal but injects back some noise.

If you take bunch of bias subs - they won't have the same value - otherwise, we would only need single bias sub. So every bias sub has bias signal - which is always the same and read noise - which is always different and mostly random (in fact even FPN is random but has different distribution to regular gaussian of read noise).

When we stack bias subs to create master bias - we are in fact trying to average out this noise - to reduce it and it works as any stacking does - SNR goes up by factor of SQRT(number of subs). Since we don't really have any interesting signal here - it is the noise that reduces by this factor - read noise.

If your camera has read noise of about 2e (modern cmos camera) and you stack 16 bias subs - you'll end up with master bias that has about 0.5e of noise.

Now if we subtract this master bias from regular light - we will be adding back in that 0.5e and we do it in the same way noise adds with other noise (although we are subtracting signal  - noise does not change if you change it's sign - it is always "jittering" around zero value - so it is always +/- something and it really makes no difference if it is -/+ instead) - in quadrature.

There is however difference when we subsequently stack such calibrated subs.

If there is no pixel shift between subs (no alignment - perfect guiding) then we can say that pixel at coordinates x=100, y=200 for example - all had same value of bias sub removed. This master bias value contains signal - that is fine, it will be properly removed from sub, but it also contains that residual noise of 0.5e. But this time when we subtract it - it will no longer be random it will always have the same values for every pixel having coordinates 100, 200 in the stack.

It has effectively became a constant rather than random value and when we stack our data it will not became smaller. If we "stack" constant values - nothing will happen average of 2,2,2,2,2,2 .... is simply 2 no matter how many "subs" we stack.

When we dither - we shuffle both target signal over sensor, but also "calibration files over stack" as calibration files match subs at pixel positions, but these get shifted when aligning if we dithered.

 

 

[ollypenrice]

2 hours ago, Budgie1 said:

I've used Photoshop for years with various things and I do like it but I also like having software that's on my own and PS now seems to be on a subscription service. Having only PS CS3 at the moment, things have moved on since this was released so I was looking for something else and I saw Pixinsight has a 45 days trail licence and thought I would get it a go.

For me, it seems to work quite well. Although it's still easy to over-do the processing and make something that looks too "processed". I do find it easier to create masks and find the effects more subtle than in PS, but there are things I know I can do easily in PS that I haven't found in PI yet, so it's swings & roundabouts.

Many well known astrophotographers are using Photoshop CS2 and CS3. They are fine for AP.

Masks can be used in both, though PI masks can be created with astro-dedicated tools. However, for me the key advantage of Photoshop is its Layers function and, with that, its selection tools. I don't have to spend ages trying to create a mask just where I want it: I can create a layer, modify the bottom layer globally and then use a selection tool (usually colour) to select what I want to modify in the top layer and erase it. What is more I can erase it at whatever percentage I like. If 100% is too much I go back one click, reduce the eraser percentage, and try again. Or I can expand-contract the selection, etc.  If I can't get the selection tool to identify the pixels I want to erase I can, if push comes to shove, zoom in and use the eraser freehand.) Above all I can see what I'm doing as I do it! I can also do custom noise reduction in Ps, targeting highly specific noise issues in a particular image. (Zoom into pixel scale, look for patterns in errant pixels, think about how to select them and then about how to modify them.  You'll always be able to find a way.)

I do some things in PI (DBE, SCNR green) but once I've done those to the linear image I get the hell out of there into a program which speaks my language! I don't speak PI but, clearly, some people do.

Olly

[Aramcheck]

1 hour ago, Rustang said:

So to clarify on your test, you stacked a dithered set of images with and with out Darks and the noise appeared to be less when you still added Darks? 

Correct, but I'll have another look at this at the weekend when we get back from our hol's. (No access to unstacked data at the moment).

Cheers
Ivor

[ollypenrice]

14 hours ago, Aramcheck said:

On @alacant's suggestion of stacking without Darks, you can of course try this & compare the noise levels on a background area. I tried it once & found that with Darks included in the process, my background noise with my astromodified 600d appeared to be about half that compared with the same set of Lights stacked with only Master Bias & Master Flats.

BTW I also have the problem with Dithering timing out on occasions.

Cheers
Ivor

But did you compare this with subtracting a master bias instead of a dark?

Olly

[Aramcheck]

50 minutes ago, ollypenrice said:

But did you compare this with subtracting a master bias instead of a dark?

Olly

In Pixinsight's ImageCalibration process, when calibrating the Lights I use a SuperBias file as a Master Bias (Calibrate unchecked) and a Master Dark (with both Calibrate & Optimise checked), so the comparison had the SuperBias included in both integrations. One caveat is that my Master Dark is taken from a set of 97 sec exposures, whereas I now take 180 sec lights.

I'll run some tests when I get back from holiday, to try to quantify the experiment.

Cheers
Ivor

[ollypenrice]

8 minutes ago, Aramcheck said:

In Pixinsight's ImageCalibration process, when calibrating the Lights I use a SuperBias file as a Master Bias (Calibrate unchecked) and a Master Dark (with both Calibrate & Optimise checked), so the comparison had the SuperBias included in both integrations. One caveat is that my Master Dark is taken from a set of 97 sec exposures, whereas I now take 180 sec lights.

I'll run some tests when I get back from holiday, to try to quantify the experiment.

Cheers
Ivor

I don't know what PI will do with that master bias or superbias, though. What was being suggested earlier, and is well known in DSLR imaging, is simply using a master bias as a direct substitute for a dark.

Olly

[wimvb]

Be very careful with optimising or scaling darks, even if a master bias is used. This may work well with standard ccd cameras, but there is no guarantee that it will work with cmos, dedicated astrocam or dslr.