Dynamic Range of Photographs

2004 July

Dynamic range
Tonal mapping
Photoshop CS "Shadow & Highlight"

Objects in the scene below lit by direct sunlight are much brighter than objects in shadows. The left photo is exposed for the sunlit areas, and the right photo is exposed for the shadow areas. Neither image looks on the computer monitor as the scene appeared to my eyes when the photos were taken; why is that? I did some searching on the internet and here's what I found.

Sunlit objects OK
(but shaded objects too dark)
Shaded objects OK
(but sunlit objects too bright)

Luminance levels detectable by our eyes range from starlight (say 0.00001 candelas per square meter) to bright sun (say 100,000,000 cd/m2). That's a range of 10,000,000,000,000 to 1. However, at any one time, the retinas of our eyes, film of a film camera, and sensor of a digital camera can only operate over a small part of that huge range -- you cannot see starlight when your eyes are adapted to sunlight. Our eyes and cameras use an aperature (and sunglasses or filters) to attentuating incoming light, providing access to luminance levels brighter than what their sensor can handle, but at the cost of losing simultaneous access to dimmer parts of the scene -- as illustrated by the photos above. Whenever the light level exceeds the dynamic range of the camera's CCD sensor, the sensor will either bottom out (record 'black') or saturate (record 'white').

I didn't make the effort of finding primary sources for this, but here are often-quoted typical estimates of maximum dynamic ranges of light sources (scenes):

Sun-lit scene100,000 : 1
Projected slide2,500 : 1
Projected negative 250 : 1
LCD projector200 : 1
Glossy print60 : 1
CRT display50 : 1
Newsprint10 : 1

... and here are often-quoted typical estimates of maximum dynamic ranges of light sensors:

EyePerhaps >10,000 : 1
Slide film250 : 1
Negative film? Greater than slide film, say many sources
Digital camera? Greater than film
High-end scanner? Similar to film

The maximum dynamic range of the light that can be capture by slide film is 250:1; interestingly, apparently that produces a slide which when projected can generate light levels with a range of 2500:1 (ie., the slide itself has greater dynamic range than the light range it can respond to). Apparently our eyes do the reverse when dark-adapted (ie., when in a dark room looking at slides), so projected slides look OK (see "Goldberg's rule"). (This may explain why photographing slides with ordinary film produces high contrast results. Also, this means that capturing all the levels recorded in a slide requires a scanner with a larger dynamic range than the dynamic range of a camera, and the values need to be compressed to cancel out the dynamic range expansion introduced by the slide.)

If an image is displayed on a CRT display or print, the perceived image with have a maximum dynamic range dependent on the capabilities of that medium (60:1 or 50:1, for CRT or photographic paper). Common techniques for mapping an image with a dynamic range larger than that of a display or print include 'gamma correction' (non-linear mapping), 'curves' (eg., Photoshop), or (most directly) Photoshop's Shadow & Highlight tool. Ansel Adams did it using 'zones' and darkroom dodging/burning. All are manipulations (mappings) of ranges of the source image into the light levels that can be produced by a display medium (eg., CRT or photographic paper).

So now we can explain what happened in the Mesa Verde scene above. The scene has a high dynamic range; deep shadows and bright sunlight. The dynamic range may be challenging even to an eye, but perhaps not noticed at the time because the eye can automatically adjust as we gaze at different parts of the scene. But when we take a photo, the camera's dynamic range (smaller than an eye) could be overwhelmed; there is no aperature setting that can shift the camera's limited dynamic range to a point that covers the larger range of that scene, so there is clipping to either white or black. However, we can cover the full range of the scene by taking multiple photos, one at the bright end (but losing the shadows to black) and another at the dark end (but losing the brights to white).

Let's take the 'good parts' of the two images above and merging them into one. The image below is stitched together from the two images above, using Photoshop, by dividing each image at the shadow line and then using the best half from each. This is a time-consuming manual method of capturing the light of a scene with high dynamic range by taking multiple photos while adjusting the camera's aperature so that each photo captures a different portion of the scene's high dynamic range, then manually selecting the 'good bits' from each to map them into a narrower range (that of the CRT).


Cliff Palace, Mesa Verde, Colorado, USA

The result above looks on my CRT more like the scene did to my eye, in my memory.

But here's a surprise! It turns out that the camera was able to capture the dynamic range of the scene above -- the histogram of values within the image below shows that the scene fell within the dynamic range of the camera (there are no peaks at all-black or all-white that would indicate clipping):

Thus the problem is mapping the dynamic range of the camera to the lesser range of a CRT (my CRT). The image below is created by dividing the photo at the shadow line and then applying a mapping (using Photoshop's curve tool) to shift the levels of the dark half higher. The histogram of the resulting image reflects the shift (the two left-most peaks have shifted right, into the zone best displayed by the CRT).

The dynamic range of the camera's CCD was able to capture the range in the scene. However, the image produced by the image on my CRT does not appear as the scene did to my eye when I was there -- the shadowed objects are too dark. By using Photoshop to make adjustments, I supply information about the scene not captured or computed in the sequence of camera to CRT, with an outcome that looks on a CRT like the scene did to an eye.

With that, I now have story to explain why images on my CRT display or on photographic prints don't always look as they did when I was gazing upon the original scene -- basically, my eye has a higher dynamic range than the combination of my digital camera and my display media. I look forward to cameras and especially display media with higher dynamic range.


Photoshop CS (version 8) introduced a function Adjustments->Shadow/Highlight that automates the above process. Below is the same image as above after having been processed with Shadow/Highlight. The result (image and histogram) is similar to the manually-produced image, but without the considerable amount of tedious work!

Photoshop CS3 (version 10) introduced Merge to HDR (merge to High Dynamic Range) that can merge several image which differ only in exposure level to create a single image with high dynamic range. Of course to be viewed, this HDR image then needs to be mapped to the narrower range of typical display media; CS3 offers several ways of doing that all-important mapping.

In 2009, the job of mapping a scene with high dynamic range to a narrower range was taken on by a camera. Sony's α550 DSLR camera offers 'Auto HDR', where two photos are taken in rapid succession and then merged using an HDR algorithm.

Sources:

(Corrections and improvements appreciated: email)

Jim Elder
Ottawa, Canada
homePage