What's wrong with digital SLR exposure metering?

When you take a photograph using a modern digital camera, it will display the image on its LCD screen, may flash some pixels that are over-exposed, and will also display an exposure histogram. These things help the photographer to assess whether or not the exposure is right, and to make appropriate corrections if not (assuming, of course, that there is time to do so). I explain here why I think that these displays are often misleading and inadequate.

This page is concerned with a different question: why do contemporary digital SLRs, when used in any exposure mode other than manual, so often get the exposure wrong?

I think there are basically two reasons:

Before discussing those points, it is worth considering what an ideal automatic exposure system would do.

An ideal exposure system

Ideally, we want to expose the image so that no channel of any pixel in the final image is either 'blown' (=255) or 'blocked' (=0). If that is so, then all the detail in the subject has been captured. We are free to use our image processing software later to alter the exposure if we wish (eg. to produce a high key or low key result).

This ideal exposure is, of course, only possible if the brightness range of the subject is less than the dynamic range of the camera's imaging sensor. If the brightness range of the subject is substantially less than the camera's dynamic range, we have the additional choice as to where on the exposure histogram to place the image. It would be sensible in that case to follow the "expose to the right" principle, because this preserves the smoothest gradation of tonality in the scene.

In practice, many subjects will have a brightness range that considerably exceeds the dynamic range of the camera's imaging sensor. What should an automatic exposure system do in those cases? I think there are roughly three possibilities:

  1. Expose to retain detail in the shadows, and let the highlights blow out
  2. Expose to retain detail in the highlights, and let the shadows block up
  3. Try to identify the real subject within the field of view, expose that correctly, and let shadows and highlights do what they will

Option (3) is what the "evaluative mode" of most contemporary digital cameras tries to do, often making use of the auto-focus point to discern what the "real subject" is. Surprisingly, only Olympus has provided any automatic way of doing (1) or (2).

Too few photocells

Since every pixel in the final image is, in principle, capable of being a different hue and brightness, ideally a camera's metering system should have at least as many photocells as the final image has pixels. If it has significantly fewer, then it may miss bright or dark bits in the subject that require adjustment of the exposure.

This suggests that the best way of assessing exposure would be for the camera to take a trial picture, using its imaging sensor, and then analyse it to determine the correct exposure. This approach certainly guarantees that one has enough photocells but, sadly, it fails for a different reason: the dynamic range of the camera's imaging sensor is too narrow. A typical exposure meter has a dynamic range of at least 18 stops, whereas a typical imaging sensor has a range of 10 stops or less.

The camera could take a trial exposure in this way but, if it then found that part of the subject was off the left or right hand end of the exposure histogram, it would have no reliable way of calculating by how much to adjust the exposure to achieve the ideal setting.

Contemporary digital cameras therefore use silicon photocells to measure the subject's brightness. These, I presume, have adequate dynamic range to do this job. Unfortunately, they are much more bulky than an imaging pixel and accordingly there are far fewer of them. How few there are can be seen by looking at a sample of real up-market digital SLRs:

UK availability dateCameraExposure systemEvaluative?Centre-weighted averaging?Partial?Spot?CompensationScene modes
Sept 2006Canon EOS 400D35-zone SPCYes*YesNo9%+-2 EV5
Sept 2006Nikon D80420-segment RGB sensorYes*Yes, severalNo2.5%*+-5 EV6
July 2006Nikon D2xs1005-pixel RGB sensorYes*Yes, severalNo2%*+-5 EV0
March 2006Canon EOS 30D35-zone SPCYes*Yes9%3.5%+-2 EV5
Feb 2006Olympus E33049-zone multi-patternYesYesNo2%, also hi- & lo-light based+-5 EV16
Mar 2007Olympus E51049-zone multi-patternYesYesNo2%, also hi- & lo-light based+-5 EV18
Oct 2007Olympus E349-zone multi-patternYesYesNo2%, also hi- & lo-light based+-5 EV0?
Nov 2005Nikon D2001005-pixel RGB sensorYes*Yes, severalNo2%*+- 5 EV0
Aug 2007Nikon D3001005-pixel RGB sensorYes*Yes, severalNo2%*+- 5 EV0
Sep 2007Nikon D31005-pixel RGB sensorYes*Yes, severalNo1.5%*+- 5 EV0
Sept 2005Canon EOS 5D35-zone SPCYes*Yes8%3.5%+- 2 EV0
Oct 2005Canon EOS 1D Mark IIN21 zone SPCYes*Yes13.5%3.8%*, also multi-spot+- 3 EV0
Feb 2007Canon EOS 1D Mark III63 zonesYes*Yes13.5%3.8%*, also multi-spot+- 3 EV0
Sept 2004?Canon EOS 1Ds Mark II21 zone SPCYes*Yes8.5%2.4%*, also multi-spot+- 3 EV0
Aug 2007Canon EOS 1Ds Mark III63 zone SPCYes*Yes8.5%2.4%*, also multi-spot+- 3 EV0
* = Linked to, or linkable to, the AF point.
SPC = Silicon photo cell, I presume

Looking at the third column above, you can see that the top-of-the-range Canon EOS 1D and 1Ds originally had a paltry 21 photocells - now increased in the mark III versions to 63 - whereas the more recent Nikon D200 and D2xs have a startling 1005 with the further advantage (perhaps) of measuring red, green and blue separately. Everything else has under 50 photocells. All these cameras thus have about 100,000 times fewer photocells than they have imaging pixels.

It is obvious that, with so few photocells, the camera cannot possibly assess the exposure correctly in many cases. All it can do is assume that the subject is of a certain kind, set the exposure on that assumption, and hope for the best.

Consequently, much depends upon the kind of assumptions the automatic exposure system makes (about the nature of the subject and the intentions of the photographer), and on what controls the camera offers the photographer to enable those assumptions to be influenced. This brings me to the second point...

The camera is deprived of necessary information about the photographer's intentions

The photographer knows whether he or she is taking a portrait or a landscape, an equestrian event or a motor race, a candid shot in the street or a still life in the studio, to list but a few of the photographic possibilities; the camera does not. Before the shutter is pressed, the camera cannot even see the scene (but the latest cameras, such as the Olympus E3, have a "live view" mode). Afterwards, the camera has an image but even the world's best pattern recognition software would struggle to discriminate between these possibilities, and would take hours of cpu time to do so.

So there is no practical way that the camera can recognise for itself what the photographer's intentions are. Nikon comes closest to attempting to do this. For example, the D200 brochure says "Nikon's 1,005-pixel RGB Exposure/Color Matrix Metering Sensor...evaluates brightness, color, contrast, selected focus area and camera-to-subject distance information, references the results against the expanded onboard database created from over 30,000 actual photographic scenes, and then calculates the final value instantly to deliver consistently dependable exposure". This may work - I have not used a Nikon. It certainly has a better chance of working than doing anything with a mere 50 or so photocells. But, like any complicated system, it probably comes to the wrong conclusion occasionally. Most importantly, it will be hard for the photographer to predict when the algorithm will go wrong and in what way.

It is interesting to note that consumer digicams are better placed. They have "scene modes" that the user can set to tell the camera what kind of subject is being photographed. A minimum of landscape, portrait, sports, night scene, snow scene are helpful, and many digicams provide a lot more, although it is doubtful if the others are much used.

As we move up the market towards the professional end, these scene modes are progressively removed from the specification (see right hand column of table above). Of course, these modes are removed for the good reason that it is assumed that the photographer knows what he or she is doing and will use aperture- or shutter-priority or even manual mode to set the exposure. But I have two issues with that.

Firstly, if I am using aperture- or shutter-priority modes, I am still dependent upon the camera's metering system to set the exposure. So I'm still at the mercy of its assumptions and limitations. Why deprive me of the ability to improve the camera's exposure accuracy by telling it what my aims are?

Secondly, if I am using manual mode, obviously I accept full responsibility for setting the shutter and aperture myself. But in that case, I want to be able to use the camera's built-in spot meter to help me to set the exposure. Unfortunately, only the Canon EOS 1D and 1Ds and the Olympus cameras help me to do this. All the others are rather inconvenient, and my Canon EOS 5D is particularly inconvenient (see here).

It is worth explaining the Olympus "highlight based spot" and "shadow based spot" metering modes because there is no equivalent on Nikon or Canon cameras. These are modes you can set the metering system into. In the highlight based spot mode, you can take a spot reading off 2% of your subject, press AEL (exposure lock) and it will set the exposure so that that part of the subject lies at the right hand end of the exposure histogram. Similarly, with shadow based spot metering, it sets the exposure so that the spot metered part of the subject lies at the left hand end of the exposure histogram. This spot mode setting is independent of whether you are using Program, aperture-priority or Shutter-priority, so you can use it with all these. You can also apply exposure compensation on top of that. I think it's a good idea, it's been around for some years, and I'm baffled as to why neither Nikon nor Canon has copied it. Perhaps there is a patent problem.

You can get an equivalent result to these Olympus highlight based and shadow based spot metering modes on other cameras by setting the exposure compensation to plus (for highlight) or minus (for shadow) half the dynamic range of the camera's imaging sensor. In practice you need about +-3 stops. Unfortunately, some Canon cameras do not offer more than +-2 stops of exposure compensation (see table above). For example, the Canon 400D, 30D and 5D all offer spot metering but only +-2 stops of compensation. This is a serious design mistake. Canon should answer the question "Given that Nikon and Olympus offer +-5 stops of exposure compensation, why do you only offer 2?"

In August 2007 Canon announced the EOS 1Ds mark III, and this has a new feature named "highlight tone priority". It is a custom function and therefore applies in all metering modes. It seems to be a combination of ISO speed adjustment and alteration of the tone curve. How this works if you are shooting raw and not using Canon's Digital Photo Professional raw conversion software, I do not know. Rather typically, Canon has not published any clear explanation of what it does, but it seems to me to be a case of changing other characteristics of the camera in order to compensate for the fact that it might have got the exposure wrong in the first place.

If you disagree with any of this please get in touch, using the email address below (which you won't see unless Javascript is enabled). Come to think of it, also get in touch if you agree with these views: just so I know I'm not a voice crying in the wilderness!

Peter Facey, Winchester, England
20071029 updated for new camera models; minor edits
20060826 originated