Shooting to the right (and why it can be a bad idea)
January 10, 2011 Leave a comment
Shooting to the right is practically a recommended technique for digital cameras. The theory is all over the internet, yet actual examples illustrating the stated improvement in output in the actual image are thin on the ground. You will find few, if any. Here’s why.
The percentage of digital noise in your image is reduced if your image is taken as bright as possible (‘shot-right’ or ‘exposed-right’), and this occurs if you
- shoot in RAW and
- keep your histogram to the far right as possible but without actually touching the right end of the histogram (to avoid clipping).
Your image will come out of the camera brighter than it should be, but you can fix that by reducing exposure in your RAW converter (Lightroom, etc).
There are many people who recommend shooting-right. Thomas Knoll (co-creater of Photoshop) is a fan, as you can see here. Adobe seem to recommend it as well, as can be seen in the Adobe white paper here (link opens a pdf in a new tab/window). If the concept of shooting to the right is new to you, I recommend that you read at least one of the links before proceeding.
Both the links give a good technical background to the issue (probably because both Knoll and Fraser are primarily looking at the technical imaging issues). As we will see, this means that they may be underplaying an important artistic consequence.
There are also other articles, such as the ChromaSoft Blog that tell us Shooting-right is ‘just plain wrong’, because it doesn’t actually reduce noise or for that matter make anything better (and simply altering ISO is a better bet). As we shall see, there actually is a good use case for shooting-right; it is just not applicable to noise.
In the first two links, there is one unanswered question; if shooting to the right is so important to achieve low noise, why don’t cameras do it by default? For a given aperture, shutter and ISO, the camera could easily look at the histogram and do the shift digitally via an automated exposure compensation behind the scenes, but put information in the RAW file that causes an automatic shift back the left in RAW conversion. Surely that would produce a competition killing camera feature?
Here’s why no camera manufacturer has added automated shoot-right compensation; shooting to the right has a consequence; you lose accurate colour reproduction.
Shooting to the right gives you a less noisy image (better signal/noise ratio) out of the camera, yet when you reduce exposure back to the left in post processing, your image becomes slightly washed out. This occurs because the red, green and blue components of your image do not change at the same rate as you overexpose and then underexpose the image, probably because your camera is doing all sorts of non-linear things behind the scenes (and even if your RAW converter knew about them by your choosing the most appropriate camera profile, there is no guarantee this will fix the issue). This non-linearity always causes the colours to change, and in the case of shooting to the right, the result is a more washed out photo.
Consider the image below. It is an amalgamation of three shots I took of a colour card placed in shade. Each square contains three strips;
The strip marked 0 (leftmost) is the image taken at normal exposure (and, for the sake of argument, shot at the camera’s base ISO), reproduced with no post processing. We are not shooting to the right and the lighting has been chosen so that the top 30% of the histogram is empty. This gives us a noisier image than the camera is capable of, but the colour is as accurate as the camera can get under the setup conditions.
The strip marked +1 is the same image but taken with exposure compensation set to +1 to better fill the top end of the histogram, then corrected in post production via a subtraction of 1 exposure value.
Similarly, the strip marked +2 is the image taken with exposure compensation set to +2 (making the histogram almost touch the right end of the scale), corrected in post production via a subtraction of 2 exposure values. This image would give us the least noisy shot the camera is capable of at our selected aperture, shutter, ISO and lens.
You would expect the three strips per square to show slight variation in brightness because of slight errors in exposure value between the camera and RAW converter. You would also expect to see differences in noise. Unexpectedly, what you see is a loss of colour and no change in noise!
When you shoot-right and then try to correct by exposing back to the left, your colours become washed out. Worse still, the colours are all washed out by different amounts so it is not trivial to get the colour back to an accurate state. As if that is no bad enough, the worst thing of all is that the colours most washed out are those associated with skin tones and there is no gain in noise!
This makes a kind of sense when you think about it. When we expose up then expose back down, we assume (correctly) that this is the same as increasing luminosity, then decresing it back to the same point. The problem is that this situation never occurs in the real world, and our eyes are calibrated for the real world: our eyes are calibrated for the sun, and that always changes colour with luminosity. The camera has to conform to our eyes, and is almost certainly using non-linear colour responses vs exposure.
What to do?
When you will be post processing your images
Thomas Knoll is right when he tells us that there is more information stored in the RAW file for the highlights. What he doesn’t tell us is why.
A camera stores more information in the highlights and less information in the shadows because that is where our eye sensitivities lie. We are more sensitive to noise in highlights than noise in shadows because we perceive highlights better. If we try to reduce noise by creating a brighter image, it will technically work (signal/noise ratio is better), but because our eyes are more sensitive to highlights, we see more noise even through there is less of it, so perceptually, we see the same amount of noise!
Put another way, camera manufacturers put more information towards the right of the histogram simply because we see noise (and everything else) at that end of the histogram better; if we actually shoot to the left instead of the right, there is mathematically more noise in the image (worse signal/noise ratio), but because our eyes are less sensitive to noise in shadows we still perceive the same amount of noise. You can shoot to the left or right as much as you want but your eye will not care; it doesn’t do the math and sees the same amount of noise. This occurs because the camera is specifically calibrated to reflect the eyes’s inaccuracy.
This leads to the one reason why shooting to the right is a good thing; when you are shooting an image that you will be postprocessing later in such a way to make portions of it brighter (i.e. dodging) and true colour is not a big issue.
Consider the following example; I am shooting an image of someone wearing a dark suit. Colour is not important as the overall key is dark anyway; the overall image is tonal. I shoot to the right, then expose down to the left in lightroom. The suit shows no texture, so I dodge (brighten) it. Two things happen;
When I shoot to the right and expose down, the shadows now contain more information than they would otherwise, but because my eyes can’t see into shadow, I don’t notice the difference.
When I dodge the suit to make it brighter, I start to see the additional information, and shooting right begins to work in my favour (if I did not shoot-right in this case, the suit would become muddy and indistinct when I open up the shadows; I begin to see the lack of information in the shadows).
Why don’t we just alter ISO? Because shooting to the right adds no new noise (it enhances signal/noise by increasing signal only; noise stays the same). ISO is physically an amplifier stage than sits after your sensor and it therefore amplifies both signal AND noise as well as adding its own noise (noise introduced by the amplifier stage itself). Changing ISO increases sensitivity to both signal and noise. Shooting to the right increases signal only. I believe this is actually a moot point though, as your computer screen is probably lacks sensitivity to see the difference unless you are brightening the image and doing some serious pixel peeping.
When you will not be postprocessing your images
Consider shooting at high resolutions in low light only if you actually need high resolution images. Strange as it may seem, shooting 6Mb images with a 12Mb camera (for example) results in less noise in the out-of-camera images (your camera is effectively using larger photosites in its sensor, reducing the effects of noise significantly). Something to bear in mind if you are taking ‘night out’ type images destined for social websites and 6×4 printed snaps only.
Better still, avoid shooting-right unless the image allows it (i.e. shoot a brighter image only if a brighter image is acceptable). Otherwise simply shoot for the key tones (or put another way; you don’t want to fill your image with data, you want your image to display your main subject accurately), and if that causes noise problems, get more sensitive camera equipment!
When you are shooting High ISO close to your camera’s limits
There is potentially one big edge case when shooting to the right makes a lot of sense: when you are shooting at a very high ISO because there is very little available light. Under these conditions, a number of things occur:
- You are at a high ISO (typically 1600 or higher) and the amount of noise makes your shots borderline usable especially because noise is now visible in the midtones and even highlights. Under these conditions, you may quite rightly consider trading color accuracy for less noise.
- The histogram has a large gap to the right because you are in low light conditions.
- There are no highlights in your image because of the low light conditions.
Consider the two images above. The image to the left was shot at ISO12800,at 1/125s shutter and f4.5. The image to the right is identical except it was shot with an exposure compensation of 1.3eV, a value that was chosen as it pushed the histogram up to the right edge without going over (it also decreased the shutter speed to 1/50s). I then decreased the exposure of the rightmost image by 1.3eV in post processing. I have not done anything else to either image.
What has happened? Well, the left image has better color (the surface the pen is sitting on is white, but looks bluish in the shot-right image), but the image to the right has much less noise and is actually usable, which cannot be said of the left one irrespective of color accuracy! So shooting to the right under high ISO conditions has saved us from an unusable shot by reducing noise considerably, but at the expense of color accuracy. I’d call that a good trade-off!
Why not just increase ISO for the second shot and then you don’t have to bother exposing down? Well, ISO and exposure compensation work differently. Increasing ISO amplifies everything.It gives you more signal, but it also amplifies the noise (as well as adding some amplifier noise for good measure). Increasing exposure compensation allows more light to the sensor by changing shutter speed or aperture (in our image, it left the shutter open longer). As long as the shutter is still fast enough to prevent camera shake and you don’t blow highlights, exposure compensation is better (and certainly looks visibly better) than a higher ISO, because the former increases light to the sensor and therefore increases signal only.
The rightmost image looks better, so why don’t cameras do this exposure compensation trick automatically? Because camera manufacturors are very conservative and don’t want to blow highlights. Even though you are in low light, your image may still contain spots of bright highlights (night sky stars or stage lighting at an otherwise dark rock concert venue for example), and the camera is trying to preserve them but is not clever enough to realise when your scene doesn’t contain any, so it doesn’t mess about with exposure compensation even when the opportunity arises.
Although noise will always reduce if you shoot right via exposure compensation, its worth noting that the practice will always reduce color accuracy. There is a trade-off between noise and color accuracy, and this typically occurs at high ISOs (ISO 1600 and above), where the reduction of noise is worth the reduction in color accuracy.
It’s also worth noting that cameras equiped with image stabilisation will prevent camera shake when using exposure compensaiton at high ISOs. For example, Sony cameras tend to have poorer high ISO noise, but Sony DSLRs also have in-camera stabilisation as standard. Using image stabilisation in conjunction with exposing right at high ISO can give Sony users results comparable with Canon/Nikon cameras at high ISO.
Carefully reading the histogram and knowing how ISO and exposure compensation both work can produce better results than the camera default in low light (i.e just increase ISO) as long as you know how to read the scene and expose for noise rather than highlights when the opportunity arises.
When you are shooting AVCHD video
This section added November 2013: adds extra information relating to video enabled DSLRs.
There is one very good reason to always shoot right: when you are shooting AVCHD video. AVCHD compression splits the scene into lots of 16×16 blocks and saves each block with a different compression based on how important it is to the overall frame. Think of it as splitting the image into lots of little JPEGS, each with a different quality.
The ‘how important it is to the overall frame’ bit is partly based on how dark the 16×16 block is (it also depends on how many color variations the 16×16 block contains, so nearly solid color is deemed less important). Thus, shadows get the most compression, and appear blocky if you expose them up in post (an effect called ‘macro blocking’). By exposing to the right when shooting AVCHD, you give the compressor less opportunities to over-compress shadows and give you macro blocking. It also helps with uniformly lit skin (which may macro block because the compressor encoded it as a solid color). See also this post for more tricks and tips when working with AVCHD.
Here’s an image with a dark overall key and subdued highlights. Overexposing (or for that matter, using a flash) would lose the gold and blue hues I’m trying to capture, leaving me with a yellow shell. Instead, I actually underexposed the image because that is the only way I can be sure of capturing the subtle colouring on the shell.
Here’s the histogram for the same image.
Although this histogram tells me that less than 50% of the image data contains anything other than zeroes, I couldn’t care less. By shooting for my key colours, I capture the image I set out to create. Sure, there’s a low amount of information in the dark areas, but my eye can’t see into the dark; perceptually it doesn’t matter.
I re-took the image, this time shot to the right. I then tried to get the same effect as the original in post by digitally underexposing. This is as close as I get;
The wonderful and subtle gold has gone, to be replaced by metalic yellow (of the type used on packaging to fake gold – yuck), and the earthy blue-black becomes a more detailed but less compelling brown.
I don’t know about you, but I prefer the first image!
Shooting right is useful if your key tones are low key AND you will be post processing later to bring parts of the key higher. Shooting to the right helps you in this case because it adds more information for you to open up in post processing. Colour should not be an issue in this case as the image will be largely tonal anyway.
Shooting to the right is also very useful in low light, high ISO conditions. If you can read the scene well enough, there are opportunities for forming an ISO, exposure compensation and shutter combination that will reduce noise much more aggressively than your camera could (or rather would dare) to do if left to its own settings.
Thomas Knoll is recommending shooting-right because he’s assuming your going to be using Photoshop afterward. Although you can’t perceive the extra information, post processing may bring out the enhanced detail associated with the extra information as you open up certain tonal ranges.
If you will not be processing your images after shooting and your images are within normal ISO ranges (50-400), shooting right has little effect. The data is stored in the RAW file in a form suited to the needs of your eye and you can forget the underlying data structure. You are not ‘losing any data’ by not shooting right. Concentrate instead on your composition and key.
- When shooting in RAW, the histogram can be misleading because it is sampled from a JPEG representation. The RAW file has better dynamic range (about 1/3 to a whole of a stop at the top end). I’ve ignored that complication but accounting for it could in theory add more information to your RAW file. I say ‘in theory’, but I’m sure nobody would rely on that theory when shooting at a wedding, because in practice, that blown out wedding dress may cost you significantly…
- I am shooting with a Sony Alpha a500. Certain makes of cameras may produce a worse/better result, but the issue will certainly exist with most cameras (particularly when many cameras incorporate a Sony fabricated sensor portion).
- The effects of shooting right at high ISO, and the resulting loss of color accuracy is mitigated somewhat by the fact that all cameras tend to lose color acuracy at high ISO anyway (noise has a random color, and this reduces overall color accuracy for your scene if noise dominates – i.e. at high ISO).
- Anyone wishing to explore the concept ‘Shooting for the key tones’ (a far better exposure policy than shooting-right in my opinion) is directed at Perfect Exposure by Micheal Freeman. You can see the promo for this book on YouTube here (link opens YouTube in new window/tab).