PetaPixel

Underexposing vs. Overexposing

One of the things I became very aware of during a recent road trip to Oregon is how much easier it is to salvage an underexposed photograph versus an overexposed one. When you overexpose a photograph and “blow out” the highlights in the image, those areas that were blown out to white are unsalvageable, whereas much more detail could have been preserved in areas of the photograph that appear to be pure black but are not (though noise results from salvaging these areas). An implication of this is that when taking portraits of people in strong sunlight outdoors, it’s much safer to underexpose and leave with a muddy looking photograph, than to overexpose and blow out detail in faces.

Of course, ideally you’d like to correctly expose a photograph, but if you need to guess, you should be conservative by underexposing. Here are a series of photographs that examines under and over exposure. The scene is half brightly lit, and half in the shadows. They were shot in RAW with a Canon 40D and a 24mm f/1.8 using evaluative metering and ISO 100. Aperture was set on f/5.6.

Properly exposed

As determined by the camera’s evaluative metering (shutter speed 1/160). The half in the shadows is not completely black, while the half under direct sunlight isn’t completely blown out either.

img_1402

First, we look at how the salvageable the image is when you underexpose it to various degrees.

1-stop under-exposed

Each of these photographs is corrected in Adobe camera raw by compensating for the under or over exposure. Hover over each one to view the uncorrected and incorrectly exposed photographs:

img_14031

2-stops under-exposed

img_1404

3-stops under-exposed

img_1405

4-stops under-exposed

img_1406

Notice how we were still able to recover a good amount of the underexposed brighter half of the scene, even though virtually all of the shadow area was lost to black. Also, notice how we’ve managed to preserve the colors of the leaves.

Now lets compare the results above with attempts to recover overexposed photographs:

1-stop over-exposed

img_1407

2-stops over-exposed

img_1408

3-stops over-exposed

img_1409

4-stops over-exposed

img_1410

Conclusion

From this experiment, we’ve found that we were able to salvage a somewhat acceptable photograph from an image that was under-exposed by four stops. On the other hand, a photo that was over-exposed by four stops was pretty much unsalvageable, as both detail and color were lost in the blown out areas.

The moral of the story is: when in doubt, under-expose.


 
  • http://twitter.com/cciotti cciotti

    For digital I totally agree but for E6 film an image under/overexposed by four stops is junk. If I know ahead of time the light will be tricky, I’ll shoot negatives instead of E6. Yes, some of us still love film :-)

  • http://www.petapixel.com Michael Zhang

    Ah, yes. I should have been more specific in saying I was discussing digital photography.

    Thanks for the comment!

  • adipatigusti

    I always underexpose by 1 or 2 stop. And shoot in raw format.

  • http://jennifergrigg.com/blog/ Jennifer Grigg

    Great information. Thanks for the analysis.

  • Steve

    This, of course, contradicts the more conventional wisdom of “expose to the right” espoused in many other websites and books. I think the explanation is as follows. Let's talk in terms of the histogram. First, we need to define what we mean by “properly exposed”. If the histogram is narrow enough to fit within the left and right limits with room to spare, then the exposure has some latitude. Increasing or decreasing the exposure will move the histogram right or left (respectively). So long as the shift does not compress the histogram against the left or right limits, then a levels or curves adjustment can shift it back to the point that gives the desired look (more about this later). So assume that under or over exposure means “crushing” one side of the histogram against the left or right limits. This means that the associated pixels are recorded as black (or white). While this block of pixels can be remapped (via levels or curves) they will always print (or display) as a single tone with no detail. Whether it is worse for this to happen to black or white pixels depends on the relative importance of highlight detail vs. shadow detail in the image under consideration. In general, the eye is more attracted to brighter areas, so highlight detail is usually more important. So it seems better to under than over expose (better to lose shadow detail than blow out the highlights). This is the case for the example above. On the other hand, assuming no under or over exposure (no crushed pixels), it is best for the histogram to be shifted as far to the right as possible because there are more bits as you move right, so more detail is recorded, and there is less noise.

  • http://www.petapixel.com Michael Zhang

    Hey steve,

    Thanks a lot for your very informative comment! Yes, I agree with everything you said. I think my point in this experiment was just to show that having a lot of “crushed” pixels at the left limit seems preferable to having the same amount of crushed pixels on the right.

    Ideally it should be as far right with as little crushing as possible, but sometimes in fast paced situations there's not always enough time to do this correctly =)

    Your comment definitely fills in a lot of details. Thanks again!

  • Steve

    This, of course, contradicts the more conventional wisdom of “expose to the right” espoused in many other websites and books. I think the explanation is as follows. Let's talk in terms of the histogram. First, we need to define what we mean by “properly exposed”. If the histogram is narrow enough to fit within the left and right limits with room to spare, then the exposure has some latitude. Increasing or decreasing the exposure will move the histogram right or left (respectively). So long as the shift does not compress the histogram against the left or right limits, then a levels or curves adjustment can shift it back to the point that gives the desired look (more about this later). So assume that under or over exposure means “crushing” one side of the histogram against the left or right limits. This means that the associated pixels are recorded as black (or white). While this block of pixels can be remapped (via levels or curves) they will always print (or display) as a single tone with no detail. Whether it is worse for this to happen to black or white pixels depends on the relative importance of highlight detail vs. shadow detail in the image under consideration. In general, the eye is more attracted to brighter areas, so highlight detail is usually more important. So it seems better to under than over expose (better to lose shadow detail than blow out the highlights). This is the case for the example above. On the other hand, assuming no under or over exposure (no crushed pixels), it is best for the histogram to be shifted as far to the right as possible because there are more bits as you move right, so more detail is recorded, and there is less noise.

  • http://www.petapixel.com Michael Zhang

    Hey steve,

    Thanks a lot for your very informative comment! Yes, I agree with everything you said. I think my point in this experiment was just to show that having a lot of “crushed” pixels at the left limit seems preferable to having the same amount of crushed pixels on the right.

    Ideally it should be as far right with as little crushing as possible, but sometimes in fast paced situations there's not always enough time to do this correctly =)

    Your comment definitely fills in a lot of details. Thanks again!

  • http://www.mcgyver.it/ Loris Albanese

    i’m not using “evaluate” metering when shoot, and i’m shoot always over exposed. why only 100 iso??

    The sensor of any camera is made to acquire information on the light. Why underexpose capturing noise?

  • Jacob

    Clearly yes, it is better to underexpose by 4 stops rather than to over expose by 4. But when in doubt overexpose by one or half a stop – for contrast you can burn details where there is some but you can’t dodge details back where there is none!