Wednesday 8 January 2014

Do High ISOs Make DSLRs More Sensitive To Light?

Many new astro-imagers believe that they can make their camera more sensitive to light by using a higher ISO setting. This is not surprising given that there are a vast number of articles out there that continue to spread this myth (including photography magazines and other technical publications that really ought to know better).

After yet another misleading write-up, this time from the BBC promoted by Stargazing Live (no less!), I thought I'd do my best to set out a few facts which will help you to understand what is really going on and therefore to take better images. Read on for more.

The Truth About ISO

The simple fact is that using a higher ISO does not make your camera's digital sensor more sensitive to light. Read that sentence again and memorise it. Anyone who tries to convince you otherwise is just plain wrong and I'll explain why.

ISO numbers are a standardised system of measurement of film sensitivity from the pre-digital era, where they did indeed indicate physical differences in the sensitivity of various film emulsions to light. Perhaps unfortunately, the concept was carried forward to digital photography as a sales/marketing tool to sell to a customer base transitioning from film to digital photography.

In truth, ISO numbers have no fundamental basis in the way digital sensors work.

Quantum Efficiency Is Where It's At

A CMOS (or CCD) sensor has what is known as a Quantum Efficiency (QE) which is the percentage of incoming photons of light that are converted in to electrons that can then be measured and converted to an image. A perfect sensor would have a QE of 100% at all wavelengths of light so that every incoming photon is detected and turned in to a measurable electron. In reality all sensors have a QE that is much lower than 100%, and the QE will also vary by wavelength.

A typical monochrome CCD sensor might have a QE of 40% at the blue end of the visible spectrum rising to perhaps 60 or 70% at the red and near infra-red end, whereas a typical (unmodified) DSLR sensor will have an (effective) QE of perhaps 40% for the pixels with the blue and green Bayer filters and maybe 30% for the red ones. So as a DSLR user, much less than half of the incoming light is going to be recorded in your image.

The lack of sensitivity at the red end is a further problem for DSLRs as it blocks most of the useful Hydrogen Alpha wavelength from emission nebulae, but it can be improved by removing or changing the infra-red cut filter which is put on top of the whole sensor. That infra-red filter is useful for daylight photographers as it produces more natural looking images out of the box as well as avoiding problems with blurry images or poor contrast caused by out-of focus infra-red light falling on the sensor.

There are many articles out there describing how to modify your camera to replace the infra-red filter for one that performs better in this regard. It is not a simple process, so if you're not willing to risk breaking your camera, you can also buy ready-modified cameras from some suppliers at a reasonable price, or direct from Canon (the 60Da) at a much less reasonable price. The results will be much better on emission nebulae, but still not as high a QE as a typical monochrome CCD.

Myth Busted

The important point is that each sensor pixel has a QE that was fixed at the point it was manufactured, and indeed the QE varies from pixel to pixel due to manufacturing imperfections. That QE will be further reduced when you put a filter in front of it (whether that is an interchangeable filter for a mono CCD or the fixed Bayer pattern filter in a DSLR).

Strictly speaking, the QE of the sensor pixel is the same, but the effective QE reduces since the filter blocks some of the photons before they get to the sensor.

The important point is that there is no setting on the camera which changes the underlying QE of the sensor elements. The same percentage of photons falling on the pixel will be converted to electrons regardless of any ISO (or gain) setting.

In the digital imaging world, ISO is just a different way of expressing 'gain'. Gain is simply the amount of amplification that is applied to the electrons in each pixel's electron well prior to converting the measured voltage to a number in the analog to digital convertor (ADC) on the sensor. Increase the gain (amplification) and a bigger number comes out of the A/D converter, which results in a 'brighter' image on the screen.

The marketing and product development bods decided to describe the camera's gain settings using the existing system of ISO numbers that their target market were already comfortable with as previously discussed. Hopefully you now understand that ISO in digital cameras has no bearing on their sensitivity to light.

As an aside, whilst the gain settings are designed to mimic the sensitivity increments of film ISO numbers, there is no absolute fixed reference point. One sensor's ISO 100 is not (necessarily) the same as another sensor's ISO 100. In any case, neither is a reflection of the true sensitivity of the sensor just that 'brightness' of the resulting image is similar to that of using ISO 100 film on the same scene.

Is Bigger Really Better?

Of course the marketing people realised that people always buy on the basis of the biggest numbers, so 'ISO wars' broke out. Everyone wants a bigger speedboat than the neighbours, and they also want a camera that offers ISO 25600 instead of one that only goes up to ISO 3200 or ISO 6400! In reality those really high ISOs are simply pointless multiplication of the digital values after the useful ADC process.

The cut off between amplifier gain and digital multiplication is usually somewhere in the region of ISO 1600 or ISO 3200, again your mileage will vary with model. This latter process (for astro-imagers) is not at all useful since you can stretch the image histogram in your image processing software to achieve the same result and do it with far more control as we'll see below.

Still Not Convinced?

If you look at an dedicated astronomical CCD camera, it does not have an ISO setting. It will either have a factory-fixed gain setting to give the best signal-to-noise ratio, or a user configurable gain setting which will just be some arbitrary range of numbers that makes sense to the camera/capture software creators (in which case you'll be expected to perform tests to figure out the best gain setting for your requirements).

Furthermore, if you look at your mobile phone, webcam or low-end digital camera it doesn't have an ISO setting even though it will have the same sort of CMOS or CCD sensor as a DSLR. At best you'll have a brightness and contrast setting which are used to manipulate both the sensor gain and the image histogram simultaneously in a way that is familiar to people who remember analogue TVs.

Again, contrast and brightness are just concepts carried over from the world of analogue TV for familiarity's sake (both in the webcam and digital TV worlds)!

Indeed with a typical mobile phone or point and shoot camera, the device itself will try to absolve you or all responsibility by analysing the image and coming up with the best settings it can untouched by human hand.

So What ISO Should You Use?

You should use an ISO setting which satisfies two key objectives:

  • Start with an ISO that is as close to 'Unity Gain' for your camera as possible (but never less than 1 electron = 1 ADU). This will ensure that you don't unnecessarily convert different voltages from the sensor pixels to the same brightness level in the final image. Unity Gain is the amplification setting that converts exactly one electron from a sensor pixel to one ADU (number) in the image file (I have covered ADUs before here). For Canon cameras Unity Gain is likely to be somewhere in the region of ISO 200 or 400. You will need to perform tests to figure this out for each specific camera model and there is unlikely to be an ISO value that is exactly Unity, so go for the nearest that is above 1.
  • Some ISO settings may produce a better signal to noise ratio than other ISO settings, so it might make sense to use an ISO which is higher (but rarely lower) than the one which approximates Unity Gain. The higher voltage coming out of the amplifier may result in less read-out noise and again your mileage will vary depending on the camera model.
Craig Stark describes how to determine the best ISO setting for your particular DSLR in his article Profiling the Long-Exposure Performance of a Canon DSLR. His advice (and mine based on practical experience) is that low ISOs are better for astro-imagers. I generally shoot at ISO400 using my Canon 500D.

But If I Use A Low ISO, My Images Are Too Dark!

Now to be fair to the BBC article I linked at the top of this piece, there is sometimes a good reason for (beginning) astro-imagers to use a much higher ISO than my advice suggests. If you are shooting using just the camera (with no laptop to control it) then you will have to rely on the tiny display screen on the back of the camera to preview and check your images. Using a low ISO may render the image hard to see or invisible, and you probably shouldn't wait until you get indoors to check your results so you may have to use a higher-than-ideal ISO.

My problem with the article (and so many others) is that it doesn't say that, it specifically says "A higher ISO number means more sensitivity...".  This leaves beginners with a misconception that stays with them for a very long time afterwards, and causes many to make fundamental mistakes in their imaging techniques once they become more advanced.

Depending on your target and exposure time, an image taken at a low ISO may well be dark, but that is not the same thing as saying that a dark image contains less information than a bright image (indeed the opposite is far more likely as we discussed above). It is just that you will need to do a bit more work to make it visible.

The solution to an image that is too dark is to stretch the histogram using an image processing package:

  • Most DSLR manufacturers supply software for converting their RAW images to other formats, and as part of the process you can usually adjust the histogram to make the image brighter. This may not be the best approach however.
  • A better solution is to bring the RAW image straight in to your photo-processing or astro-image processing software (e.g. PhotoShop, PixInsight, Deep Sky Stacker, etc.), all of which have tools for manipulating the histogram to make the image brighter. This will ensure that you start out with the maximum amount of detail in the image and have full control over how it is manipulated.
  • The processes in the various packages have different names such as Histogram Equalisation, Curves, Histogram Transformation, Brightness and Contrast, etc. In all cases, the software can make a seemingly 'black' and very unpromising astro-image reveal the hidden glory of stars, galaxies and nebulae.
In saying all this, I've assumed that you are shooting RAW images with your DSLR rather than JPEG images. If you aren't, please change your settings now! JPEG images throw away a lot valuable detail in an image (both in terms of brightness levels and resolution). As an astro-imager you should always shoot RAW images and only ever convert them to JPEG or other web-friendly formats at the very end of processing.

The Bottom Line

As an astro-imager, really you need to focus on gain (through the imperfect medium of ISO numbers for DSLR users) and histogram transfer functions (in post-processing) to avoid misunderstanding what it is you are really doing when creating or processing an image.

No comments:

Post a Comment