Bias frames and image calibration

[MartinB]

Most imagers know about darks and flats. Darks to remove hot pixel noise from dark current and flats to remove gradients, vignetting and dust artefacts. However the role of bias in both light and calibration frames is sometimes overlooked. This is a little primer to for anyone who might be having problems with good image calibration.

What is bias

Look at a plain sheet of white paper. What appears to be a uniform white is on closer inspection a varied collection of shades giving a mottled appearance. This is just the same as your CCD chip. Unstimulated pixels produce a small variation in brightness value. In fact, to avoid pixels reporting negative values all the pixels in the chip are given a small "boost". This might be around 200 ADUs (analogue to digital unit - or brightness levels in plain English), which might have your unexposed pixels reporting levels of between 180 and 230.

Does bias matter?

Yes - no surprise there! In the same way as a sheet of paper appears plain white ordinary photos show no evidence of bias. That's because the pixel variation due to bias is very small and at very dark values. However, when we stretch the image this starts to change. As you do your histogram stretch your target will start to emerge like magic. Unfortunately as you start to get towards a satisfactory stretch the background turns nasty. Even ignoring hot pixels, gradients and dust motes the background is mottled and unpleasant. Processing this out is very difficult and compromises the final image. This is the effect of the bias, the good news is that it can be removed.

How to take a bias frame

Fortunately this is very easy. An ideal bias frame should have no exposure at all. Some cameras allow this but others require a minimum exposure, this isn't really a problem, just make sure the camera is fully protected from bright light. As always, take a few frames and then combine them to create a master bias.

Using bias frames with flats and darks

This is where all the trouble starts!!! Your darks and flats contain the same bias as you light frames. It is very easy for inappropriate bias adjustments to be made during image calibration.

Bias subtraction with darks

A perfect dark will have the same exposure as your light frames with the CCD at the same temperature. If you have darks like this just apply them in the normal way and not only will the dark current be subtracted from the light but the bias will be taken out as well. Simple!

Scaled darks

Getting ideal darks is pretty straight forward with a temperature controlled CCD. You can collect the darks at any time. This isn't an option with DSLRs and CCDs with uncontrolled cooling such as an SXV H9. It seems a crying shame to sacrifice light frame exposure time to collect darks. Fortunately it is possible to take darks of shorter exposure than the light frames and "scale" them. So, if you light frames are 500 secs you could take 125 second darks and scale them by multiplying the pixel values by a factor of 4. However, bias is constant, you don't want to be multiplying that by 4!

If you are using scaled darks you need to bias subtract the dark before scaling. Your image calibration software should do this for you without any problem. Once you have scaled you will have a reasonable dark which can be subtracted from the light frame however, in this situation the light frame will still contain the bias. Just subtract the bias after subtracting the dark.

Flats and bias

Your flat frames contain the bias. Applying a bias containing flat frame to a bias containing light frame DOES NOT remove the bias, it messes things up! A flat frame is not subtracted from the light frame, the maths is different. You should always subtract the bias from your flat frames. Most calibration software will recognize "bias subtracted flats" as being distinct from uncalibrated flats. Of course, if you want to, you can apply appropirate darks to your flats and this will also remove the bias. Whatever you do, don't calibrate the flats with darks and then bias and end up subtracting the bias twice!

Summary

Bias subtraction is important to achieve a smooth background in your final image. Collection of a good bias frame is easy but care needs to be taken during calibration to ensure bias is subtracted appropriately.

I hope this is of some help

[jgs001]

Thank you Martin, very useful indeed. So even if your darks are taken at the same time, in what should be the same conditions, with the same exposure time, if the temperature has varied in any way (it probably has cooled somewhat) and the camera may have got warmer after capturing a load of lights anyway, it's better to use bias to adjust the darks. This also explains why, the only time I tried using bias frames the image got messed up. The bias and flat darks combined did that.

[MartinB]

Unless you have controlled cooling the darks are always going to be an approximation. The most accurate method would be to intersperse light collection with dark collection. You only want to bias adjust the darks if you are going to scale them.

[jgs001]

Thanks Martin. One more... If the only factor in bias frames is the ISO, is keeping a store of bias frames a workable option. In fact, keep a store of Master bias frames, one for each ISO setting ?

[Sam]

Martin,

This is very useful information, thanks heaps for posting it. I have a question about this and DSS, specifically. I take my lights in RAW, my darks in RAW and my bias in RAW (I haven't got to flats yet). When I stick them all into DSS does DSS automatically subtract the bias from the darks wihtout me telling it to? or does it just subtract the bias from the lights? and then subtract the unbiased (is that a word?) darks from the lights? or does it automatically know to subtract the bias from the lights and darks before subtracting the darks from the lights?

cheers

Sam

[Dave Smith]

Thanks Martin, this is very instructive and has sorted one puzzle in my mind. I am very interested in your answer to Sam (lightbucket).

Dave

[Psychobilly]

Martin : Another Exellent "primer"

Sam : Yep Dss sorts it all out for you...

John : I Take some Darks at the Start of the session ... some in the Middle and some at the End... and look to see if there are any significant changes between them - diff them in CS...

Billy...

[jgs001]

Sam, I just found this in the DSS help. I've finally got around to reading the docs more .

"Creation of the final picture by adding all the light frames with the selected method.

The master offset and the master dark are automatically subtracted from each light frame and the result is divided by the calibrated master flat, then if the option is enabled the hot pixels detected in the dark frame are removed and the value is interpolated from neighbors.."

[MartinB]

Thanks Martin. One more... If the only factor in bias frames is the ISO, is keeping a store of bias frames a workable option. In fact, keep a store of Master bias frames, one for each ISO setting ?

Never used a DSLR in my life John but if the iso acts in the same way as gain then, yes, store a set of master biases for each iso setting.

I don't use DSS but have had a look at the help files and, as Billy says everything is done automatically. It refers to bias as offset which can get confusing! It creates a master bias and then subtracts this from each dark frame and creates a bias subtracted master dark. The master offset and master bias are then subtracted from the individual light frames. Below is taken from the help file.

Step 5

Creation of the final picture by adding all the light frames with the selected method.

The master offset and the master dark are automatically subtracted from each light frame and the result is divided by the calibrated master flat, then if the option is enabled the hot pixels detected in the dark frame are removed and the value is interpolated from neighbors..

I don't know how you go about scaling darks in DSS

[jgs001]

Thanks Martin. To scale, there's an option in the Dark stacking parameters, dark multiplication factor, I guess it's that.

[MartinB]

That sounds like it John

[Psychobilly]

Its worth getting to "know" your cameras noise characteristics......

With the older DSLR's which suffered significant ampglow you could watch the ampglow increasing frame to frame..

Whenever I get a new DSLR I always try and characterise its performance y...

Taking Darks of increasing length (double each time) with a significant gap between them at the various ISO settings and looking at them closely in CS... OK time consuming but need to know how they perform...

Next pick a reasonable longish exposure and take successive darks (say 5-10) with a gap between them and see how the dark varies frame to frame... start with along gap and then reduce it (usually half it each time) and see what effect this has.. Your looking for timings which give you consistent darks with little or no variation between them..

As an example my Nikon D200 needed typically 10 frames to reach "equilibrium" and needed a gap of at least half the length of the light frame between successive lights...

The 350D was better and is significantly better after removing the IR filter...

My Canon 1000D has no significant ampglow even with 20 min exposures at room temp and I typically only leave 15s between exposures...

Billy...

[roundycat]

Martin, one comment if I may; Scaled darks should be a longer exposure than the light frame you are subtracting the dark from. In other words, you can use a 10m scaled dark to calibrate a 5m light but not the other way round. In a sense one method is interpolation which is safe but the other method (10m light, 5m dark) is extrapolation which is not safe.

A description I heard from Ron Wodaski is that the bias is what your camera does at the start of the exposure, the dark frame shows what is happening at the end of the exposure. Of course we don't want either in the picture.

Dennis

[MartinB]

Well Dennis I've read around this a fair bit. I have heard that Richard Berry advocates darks of x5 the light exposure time and then scaled down. Others (inc Ron Wodaski) say that gives them a very poor result.

There is no powerful reason why you can't scale up and I have read comments from Ron W himself suggesting this is a reasonable thing to do. The results wont be perfect because pixel response to dark current isn't completely linear over time, that is the main limitation for scaling darks. You will always get a few extra hot and cold pixels.

The best bet is for people to try scaling with their own kit and find out what works and what doesn't. If scaling up gives a reasonable result then that could result in them gaining a whole lot of extra light exposure time. keep all the master darks and build a library over the years (that would work except we keep changing cameras!)

[ngc2403]

Excellent Primer Martin, Really well written and clear!

Thats the way i do mine and the way i have read to do them

thanks though for putting it so clearly

[Taff]

Thats a nice write up Martain' on the subjuct time to read and make it sink in.Mark