11.2.1 What we don't know
DSLR manufacturers do not release detailed specifications of their sensors. Accordingly, the sensitivity curves in Figure 11.5 (p. 134) reflect a certain amount of guesswork.
What's more, the "raw" image recorded by a DSLR is not truly raw. Some image processing is always performed inside the camera - but manufacturers are extremely tight-lipped about what is done. For example, Canon DSLRs apparently do some kind of bias frame subtraction on every image, and Nikon's "star eater" speckle-removing algorithm is notorious.
We can hope that in the future, DSLRs will come with the equivalent of film data sheets, giving at least the spectral response, characteristic curve, and signal-to-noise ratio. Until that happens, we have to rely on third-party tests.
For detailed investigations of astronomical DSLR performance, see the web sites of Christian Buil (www.astrosurf.net/buil) and Roger N. Clark (www. clarkvision.com). Buil focuses on astronomical performance and spectral response, while Clark sets out to measure sensor parameters such as full-well electron capacity and signal-to-noise ratio.2
Another way to test a DSLR is to ignore the internal details and use well-established measures of picture quality, such as dynamic range and color fidelity. This is the approach taken by the Digital Photography Review web site (www.dpreview.com) and by the reviews published in European magazines such as the British Journal of Photography, Chasseur d'Images (France), and Super Foto (Spain).
All of these tests point to one important result: DSLRs are improving steadily and rapidly. The latest models have more dynamic range and less noise than those even two years older, and all current DSLRs perform much better than film.
The pixels in a DSLR sensor are much larger than those in a compact digital camera. That's why DSLRs perform so much better. Typically, a DSLR pixel is 5-8 |im square and can accumulate over 40 000 electrons. Besides the desired signal, a few electrons always leak into the cell accidentally at random, but each of them will constitute only 1/40 000 of the total voltage. In the smaller pixels of a compact digital camera, the same electrons would do much more harm.
2 Clark is also the author of Visual Astronomy of the Deep Sky (Cambridge University Press, 1990, now out of print), a groundbreaking study of the factors that determine the visibility of faint deep-sky objects to the human eye.
Each cell has to contain an integral number of electrons. You can have 23 624 electrons or 23 625, but not 23 624^ because there's no such thing as half an electron. This is why the camera cannot distinguish an infinite number of different brightness levels. But then, neither can light itself; there's no such thing as half a photon, either.
A much coarser kind of quantization takes place at the output of the amplifier, when the cell voltage is turned into a digital signal. Normally, DSLRs perform 12bit digitization, meaning that each cell voltage is rendered into a whole number between 0 (black) and 4095 (white). This number is of course less precise than the actual electron count.
These 12-bit numbers are often called ADUs (analog-to-digital units). They can be stored in a 16-bit TIFF file, where they do not use the whole available range. Such a file, viewed directly, looks very dark until it is "stretched" by multiplying all the values by a constant. Even then, it may look rather dark because it needs gamma correction (p. 170).
To provide adjustable ISO speed, the camera lets you amplify the voltages during the digitization process. That is, you don't have to use the full capacity of the cells. Suppose your cells hold 50 000 electrons. You can multiply all the output voltages by 2, and then 25 000 electrons will be rendered as maximum white (and so will anything over 25 000). This means you need only half as much light, but the image will be noisier because every unwanted electron now has twice as much effect.
The dynamic range of a sensor is the range of brightness levels that it can distinguish, usually measured in stops, where (as in all photography) "N stops" means a ratio of 2N to 1. Tests by Roger Clark show that, in raw mode, DSLRs can cover a brightness range of 12 to 15 stops (that is, about 4000:1 to 30 000:1), which is appreciably greater than the 9- or 10-stop range of film. However, when pictures are output as JPEG, the usable dynamic range is no more than 8 or 9 stops. The rest of the range is used for ISO adjustment.
DSLRs normally have the largest dynamic range at ISO 100 or 200, as well as the best signal-to-noise ratio. The earliest DSLRs got steadily worse as you turned up the ISO speed. But tests by Roger Clark, as well as my own experience, show that with newer DSLRs, there is very little difference up to ISO 400. After that, compromises become evident, but higher settings can still be justified when the object being photographed is faint.
Color balance is similar to ISO speed adjustment except that the red, green, and blue pixels are adjusted by differing amounts. This corrects for the inherently uneven sensitivity of a CCD (stronger in red than in green or blue) and allows for variations in the light source. The two obvious choices for astrophotography are daylight balance, to get the same color rendition all the time, or automatic white balance, to try to avoid a strong overall color cast. Color balancing is done after digitization and only affects JPEG images, not raw image files. In software such as MaxDSLR, you get to specify multiplication factors for the three colors yourself.
Every sensor has a few flaws. Hot pixels are pixels that always read out as an excessively high number due to excessive electrical leakage (dark current); dead pixels always read out as zero.
Of the two, hot pixels are more of a problem. You can see them by making a 5-minute exposure with the lens cap on. Most of them will look bright red, green, or blue because they hit only one spot in the Bayer color matrix. Indeed, vividly colored stars in a DSLR photograph should be viewed with suspicion; they may not be stars at all.
In general, the hot pixels in a sensor are reproducible; they will be the same if you take another exposure soon afterward. That's why dark-frame subtraction is effective at eliminating them.
Even among the pixels that are not hot or dead, there is inequality. No two pixels have exactly the same sensitivity to light, the same leakage, or the same bias. Bias is the starting voltage, non-zero because you can't get all the electrons out of every cell before starting an exposure. Reproducible inequality between pixels is called fixed-pattern noise. Besides appearing as random grain, it may also have a striped or tartan-like pattern.
With modern DSLRs, fixed-pattern noise is usually very slight and is strongly corrected by the computer inside the camera.
Leakage, or dark current, is affected by temperature; that's why astronomical CCD cameras have thermoelectric coolers. Even DSLRs are noticeably less noisy in the winter than in the summer; that's why dark frames should always be taken at the same temperature as the images from which they will be subtracted.
Theoretically, the dark current of a silicon CCD sensor doubles for every 8° C (15° F) rise in temperature.3 This relationship is affected by the way the sensor is fabricated, and I have not investigated whether it is accurate for the latest
3 Extrapolating from the curve on p. 47 of Steve B. Howell, Handbook of CCD Astronomy, 2nd ed. (Cambridge University Press, 2006), to normal DSLR camera temperatures. At lower temperatures the change per degree is greater.
Figure 11.3. Amplifier glow (arrow) mars this image of the Rosette Nebula. Single 10-minute exposure at ISO 800, unmodified Nikon D50,14-cm (5.5-inch) f /7 TEC apochromatic refractor. Dark-frame subtraction was later used to remove the amp glow. (William J. Shaheen.)
DSLR sensors. What is definite is that there's less noise in the winter than in the summer.
At this point you may be thinking of chilling the camera. This has been tried, but one obvious drawback is that moisture from the atmosphere will condense on the sensor if the sensor is cold enough.
Blooming is a phenomenon that makes an overexposed star image stretch out into a long streak. It is uncommon with DSLRs though common with astronomical CCDs. Blooming occurs when electrons spill over from one cell into its neighbors.
Amplifier glow (electroluminescence)
The main amplifier for the CCD or CMOS sensor is located at one edge of the chip, and, like any other working semiconductor, it emits some infrared light. Some sensors pick up a substantial amount of "amp glow" in a long exposure (Figure 11.3). Dark-frame subtraction removes it.
Even if the sensor is perfect, pixels will occasionally be hit by ionizing particles from outer space. These often come two or three at a time, byproducts of collisions of the original particle with atoms in the air (Figure 11.4). Like hot pixels, cosmic ray hits are likely to be vividly colored because of the Bayer matrix.
Cosmic rays are a source of non-reproducible hot pixels. They are also the reason you should not believe just one digital image if it seems to show a nova
or supernova. When conducting a search for new stars, take every picture at least twice.
A simple way to reduce noise and remove flaws is to bin the pixels, i.e., combine them in 2 x 2 or 3 x 3 squares. This results in a picture with, respectively, 1/4 or 1/9 as many pixels as the original.
It takes about a million pixels to make a pleasing full-page picture. This means the output of a 10-megapixel DSLR can be binned 3 x 3 with pleasing results.
The easiest way to accomplish binning is simply to resize the image to 1/2 or 1/3 of its original linear size using Photoshop or another photo editing program. Many astronomical software packages offer binning as an explicit operation.
Was this article helpful?
Compared to film cameras, digital cameras are easy to use, fun and extremely versatile. Every day there’s more features being designed. Whether you have the cheapest model or a high end model, digital cameras can do an endless number of things. Let’s look at how to get the most out of your digital camera.