What are Bias frames

[Frank the Troll]

Ok before I dive in, help out a newbie.  I get lights, flats & dark frames, but what are bias frames

[DaveS]

Bias Frames are like Dark Frames, but taken at the shortest possible exposure, to reveal any underlying unevenness in the sensor. For CCD I use Bias but not Dark, for CMOS Dark is vital but Bias can stuff up your stack.

[mark skelton]

Bias frame are like dark frames ie with lens cap on but the exposure is set at the fastest exposure ( in case of my dslr 1/4000 sec.) where as dark are the same exposure length as lights.

[david_taurus83]

Typically not used with the popular ASI1600 as the ADU count is small. But on a DSLR the bias offset is quite large. On my 6D it is 2047 ADU at ISO1600 out of a possible 16k ADU. So a must for me.

[Merlin66]

Why take bias frames?????

There are only of use when scaling darks for different temperatures????

All images include bias.

Flats and Flat darks should be all that is ever required.

Agree/ disagree??

 

[david_taurus83]

1 minute ago, Merlin66 said:

Why take bias frames?????

There are only of use when scaling darks for different temperatures????

All images include bias.

Flats and Flat darks should be all that is ever required.

Agree/ disagree??

 

I use a master bias to calibrate my flats. I have a temp matched dark library for my lights. Saves me shooting flat darks each session as flat exposures tend to vary.

 

But your method described above was exactly how I used to work with my ASI1600.

[Frank the Troll]

31 minutes ago, DaveS said:

Bias Frames are like Dark Frames, but taken at the shortest possible exposure, to reveal any underlying unevenness in the sensor. For CCD I use Bias but not Dark, for CMOS Dark is vital but Bias can stuff up your stack.

Thank you.  So can I assume, they are not really needed when using a DLSR, or am I missing something?

Edited January 22, 2020 by Frank the Troll

[alacant]

56 minutes ago, Frank the Troll said:

not really needed when using a DLSR

I find they are essential. They take only a short time to produce and once you have them, you can use them over and over.

You can however safely dispense with dark frames -for my eos they introduce more noise- by taking only bias and flat frames but dithering between light frames. Most modern stacking software will produce cleaner results with DSLR data this way.

Of course YMMV but HTH anyway.

 

[dph1nm]

Both CMOS & CCD manufacturers tend to add a constant value to the image before it is read out, in order to avoid negative values. This needs to be subtracted (both from flats and lights) before flats will work correctly. Easiest way to find out what this value is is to take zero-length exposure (i.e. with no light getting in), known as the bias. For DSLR cameras the best approximation to 'zero-length' is to use the shortest shutter speeds available with the camera in the dark.

You could use darks and flat-darks instead, as darks also have the bias signal added, so subtracting a dark gets rid of bias as well.

NIgelM

 

[Pompey Monkey]

1 hour ago, dph1nm said:

Both CMOS & CCD manufacturers tend to add a constant value to the image before it is read out, in order to avoid negative values. This needs to be subtracted (both from flats and lights) before flats will work correctly.

 

They do indeed add a constant value to the image to avoid negative numbers. However, this is a constant and is called the pedestal. It is not the bias signal.

The bias is electronic noise generated by the sensor and camera electronics as the image is read off the chip. Fortunately, this noise is fairly consistent for every image read and, by averaging many bias files, a model can be built ("master bias") that can be subtracted from every image to dramatically reduce the read noise from the camera/sensor.

20 hours ago, Merlin66 said:

Why take bias frames?????

There are only of use when scaling darks for different temperatures????

All images include bias.

Flats and Flat darks should be all that is ever required.

Agree/ disagree??

 

Bias should be subtracted from every image read from the camera i.e. lights, flats and darks. Except bias frames, of course

[ollypenrice]

23 hours ago, Merlin66 said:

Why take bias frames?????

There are only of use when scaling darks for different temperatures????

All images include bias.

Flats and Flat darks should be all that is ever required.

Agree/ disagree??

 

Disagree with CCD imaging to the extent that you could, if you wished, faff about taking darks-for-flats but it really is a total waste of time since you can use a master bias as a dark for all your flats and find no difference whatever. This is not the case for CMOS flats which need 'correct' darks for flats.

I take it one step further, but in doing so I 'fess up as a cheapskate. I use a luminance flat to calibrate all my stacks in whatever filter. Repeated comparisons have shown that this makes no difference whatever most of the time. On the very rare occasions when it does make a difference I'll shoot dedicated flats 'per filter.'

Olly

[Stub Mandrel]

Just don't do what I did this week and use Sharpcap's automatic dark subtraction. And subtract your master dark flat from all your lights...

[vlaiv]

4 hours ago, Pompey Monkey said:

Bias should be subtracted from every image read from the camera i.e. lights, flats and darks. Except bias frames, of course

That really depends. I most cases you in fact don't need bias to be subtracted and if you follow "standard" workflow you can actually use any file as bias - even Picasso painting digitized to exact size as your subs

Let me explain and show why is that:

If we observe "regular" calibration procedure being (same thing happens to flats, so we will skip flat calibration for now, and just mention it at the end):

- Master bias is stack of bias subs (for now, we will later substitute in Picasso painting instead)

- Master dark is made by stacking "calibrated" dark subs.

- Calibrated dark sub is dark sub minus master bias

- Calibrated light =  (light - master bias - master dark) / master flat

Let's do a bit of substitution

Calibrated light = (light - master bias - average(dark - master bias) ) / master flat

Now average is regular average sum and division, and if we have "constant" term we can pull it in front of the brackets so let's do that

Calibrated light = (light - master bias - (average(dark) - master bias) ) / master flat.

Let's rearrange that a bit:

Calibrated light = (light - master bias + master bias - average(dark) ) / master flat

Now you will probably notice that we have -master bias and +master bias and fact is that two numbers with same absolute value - one negative and one positive added together will give 0, and we can have any old "number" there it won't change a thing so it is safe to also write this:

Calibrated light = (light - Picasso image + Picasso image - average(dark)) / master flat

and that is equal to

Calibrated light = (light - average(dark)) / master flat

You don't need bias to do proper calibration, and in fact if you use above "standard" calibration flow - you can use any, and I mean literally any sub as master bias - it will make no difference at all.

You only need bias in very special cases - like mentioned above by @Merlin66 - scaling darks - either in form of different exposure length or when trying to optimize dark calibration (darks at different temperature).

[ollypenrice]

10 minutes ago, Stub Mandrel said:

Just don't do what I did this week and use Sharpcap's automatic dark subtraction. And subtract your master dark flat from all your lights...

Well, I do my stacking in AstroArt which has a pretty simple format. Sometimes I lie to it, so I'll put a set of flats into the 'Images' box and put a master bias into the 'darks for images' box and ask it to 'average combine' this lot into something I'm going to save as 'Master flat.' It works to my satisfaction because I know what the software is doing. And that's the key thing: understand the steps. I know you do. I think the best advice we can give to beginners is, 'Think it through.'

Olly

[Pompey Monkey]

1 hour ago, vlaiv said:

That really depends. I most cases you in fact don't need bias to be subtracted and if you follow "standard" workflow you can actually use any file as bias - even Picasso painting digitized to exact size as your subs

Let me explain and show why is that:

If we observe "regular" calibration procedure being (same thing happens to flats, so we will skip flat calibration for now, and just mention it at the end):

- Master bias is stack of bias subs (for now, we will later substitute in Picasso painting instead)

- Master dark is made by stacking "calibrated" dark subs.

- Calibrated dark sub is dark sub minus master bias

- Calibrated light =  (light - master bias - master dark) / master flat

Let's do a bit of substitution

Calibrated light = (light - master bias - average(dark - master bias) ) / master flat

Now average is regular average sum and division, and if we have "constant" term we can pull it in front of the brackets so let's do that

Calibrated light = (light - master bias - (average(dark) - master bias) ) / master flat.

Let's rearrange that a bit:

Calibrated light = (light - master bias + master bias - average(dark) ) / master flat

Now you will probably notice that we have -master bias and +master bias and fact is that two numbers with same absolute value - one negative and one positive added together will give 0, and we can have any old "number" there it won't change a thing so it is safe to also write this:

Calibrated light = (light - Picasso image + Picasso image - average(dark)) / master flat

and that is equal to

Calibrated light = (light - average(dark)) / master flat

You don't need bias to do proper calibration, and in fact if you use above "standard" calibration flow - you can use any, and I mean literally any sub as master bias - it will make no difference at all.

You only need bias in very special cases - like mentioned above by @Merlin66 - scaling darks - either in form of different exposure length or when trying to optimize dark calibration (darks at different temperature).

Wow! that's a lot of explanation. Thanks.

The way my, rather limited, interpretation is if you subtract the bias (and dark if necessary on long exposures), then:

• Flats (Vignetting/dust bunnies) are a multiplicative correction factor,
• Gradients are subtractive.

Yes/no?

[vlaiv]

9 hours ago, Pompey Monkey said:

Wow! that's a lot of explanation. Thanks.

The way my, rather limited, interpretation is if you subtract the bias (and dark if necessary on long exposures), then:

• Flats (Vignetting/dust bunnies) are a multiplicative correction factor,
• Gradients are subtractive.

Yes/no?

Ok, now I'm confused

We calibrate our subs so all that remains is light signal. In this process flats are always multiplicative factor and should be applied only to light signal as light is only thing that is affected by blockages and shadowing.

If we look at simplified model, raw sub that comes out of the camera contains bias signal, dark signal, light signal (here we don't discriminate between target and sky and we yet have no idea of vignetting / dust).

We need to remove bias and dark so only thing that is left is light signal. We do this by taking darks. Raw dark contains all but light as scope aperture is covered when taking these subs - all light is blocked. This means that raw dark subs contain both bias signal and dark current signal. If we subtract this from our lights we have done the job (this is my explanation above, no need to fiddle around with bias here for simple calibration).

Once you have only light signal left, the you can correct multiplicative factor of blockage / shadows by dividing by master flat (that also needs to be made out of "pure light" with all other signals removed).

Gradients are quite special - it is guess work rather than calibration. In principle you can't tell if LP gradient is coming from target or not. Both are light and you can't distinguish how many photons belong to sky and how many to the target. You can only guess by using some sort of approximation - like sky is either constant or linear gradient (or maybe simple polynomial of certain degree) and you can't have negative values, but you know that certain parts of the image don't contain target.

So answer to above question would be something like: Vignetting / dust is always multiplicative but for flat division to work properly you need proper removal of all signals but light signal. Gradients are additive/subtractive and in general don't depend on proper calibration - neither proper calibration helps with their removal nor you need proper calibration to attempt to remove them.

Did I get your question right?

[Pompey Monkey]

Yes you did 😃

I agree that subtracting the uncalibrated darks also removes the bias from the lights.

I just prefer to remove the  bias from all my files so I know that I have "a level playing field" from the start. 😃

[dph1nm]

21 hours ago, Pompey Monkey said:

They do indeed add a constant value to the image to avoid negative numbers. However, this is a constant and is called the pedestal. It is not the bias signal.

When I started CCD imaging many,many years ago this is exactly what was done. An average value was calculated from the bias frames (or indeed the overscan region)  and subtracted from the lights and flats. We never subtracted bias frames to avoid adding noise. It worked perfectly well.

NigelM

[Stub Mandrel]

19 hours ago, vlaiv said:

That really depends. I most cases you in fact don't need bias to be subtracted and if you follow "standard" workflow you can actually use any file as bias - even Picasso painting digitized to exact size as your subs

Let me explain and show why is that:

If we observe "regular" calibration procedure being (same thing happens to flats, so we will skip flat calibration for now, and just mention it at the end):

- Master bias is stack of bias subs (for now, we will later substitute in Picasso painting instead)

- Master dark is made by stacking "calibrated" dark subs.

- Calibrated dark sub is dark sub minus master bias

- Calibrated light =  (light - master bias - master dark) / master flat

Let's do a bit of substitution

Calibrated light = (light - master bias - average(dark - master bias) ) / master flat

Now average is regular average sum and division, and if we have "constant" term we can pull it in front of the brackets so let's do that

Calibrated light = (light - master bias - (average(dark) - master bias) ) / master flat.

Let's rearrange that a bit:

Calibrated light = (light - master bias + master bias - average(dark) ) / master flat

Now you will probably notice that we have -master bias and +master bias and fact is that two numbers with same absolute value - one negative and one positive added together will give 0, and we can have any old "number" there it won't change a thing so it is safe to also write this:

Calibrated light = (light - Picasso image + Picasso image - average(dark)) / master flat

and that is equal to

Calibrated light = (light - average(dark)) / master flat

You don't need bias to do proper calibration, and in fact if you use above "standard" calibration flow - you can use any, and I mean literally any sub as master bias - it will make no difference at all.

You only need bias in very special cases - like mentioned above by @Merlin66 - scaling darks - either in form of different exposure length or when trying to optimize dark calibration (darks at different temperature).

Ah, the place where bias (allegedly) belongs is master flat = (average flats)-master bias.

Unless you use master flat = (average flats)-master dark flat