Doug Kerr
Well-known member
There is (quite properly) considerable interest in the dynamic range capabilities of digital cameras (an as well of techniques for enhancing the effective dynamic range through image stacking and the like).
But when we begin comparing numerical expressions of the dynamic range of different cameras, we need to be mindful that there can be various ways to "score" that property, and not always will the same one have been used for the various "competitors".
The basic concept of course is straightforward: we consider the dynamic range of a camera (or imaging chain) to be the ratio of the largest and smallest luminance that can exist in a scene such that (in a single exposure) all the detail (of interest) of the scene is "usably captured".
It is the matter of "usably captured" that, in part, distinguishes the different possible quantitative definitions of dynamic range.
Note that when we say "luminance", we usually mean "base luminance" (my term), the "average" luminance of a scene area in which detail exists in the form of (small) variations in luminance above and below it.
One possible definition would be this:
The maximum luminance to be considered in the ratio is the highest one at which the slope of the exposure result curve (the digital equivalent of the "D log E" curve for film) is at least 90% of its value at a luminance a 1/2 stop lower.
The minimum luminance to be considered in the ratio is the highest one at which the slope of the exposure result curve is at least 90% of its value at a luminance 1/2 stop higher.
In effect, we consider the limits of the luminance range to be where "clipping" of detail has begun to emerge to a certain degree. (One could of course use different numerical criteria; my numbers were just to illustrate a type of definition.)
A comparable definition is one often used in connection with film.
But note that the value for "dynamic range" obtained under such a definition will greatly depend on the choice of those numbers in the definition.
However, some may argue that just because the imaging sensitivity retains its "incremental sensitivity" (a reasonable interpretation of the slope of the curve) at a certain small base luminance does not mean that we get a useful reproduction of detail about that luminance, because of the intrusion of noise.
Thus, we sometimes use a definition of dynamic range in which the "lower luminance" used in the ratio is the lowest luminance at which the "equivalent signal-to-noise ratio" is at least a certain stated target value (perhaps 1:1). (I say "equivalent" because of a subtlety that makes actual single-to-noise ratios as small as the usually-stated bogey impractical to measure directly.)
The ISO definition of the dynamic range of a digital camera is predicated on that concept for the lower luminance. In it, the "upper" luminance has a very simplistic definition: the luminance at which "sudden clipping" of the digital result occurs (as a result of exhausting the range of the numbering system used).
A further very important issue is where do we examine the "digital exposure result" to discern what scene luminances qualify as the limits of the luminance range (the numerator and denominator of our ratio). In the ISO definition, it is the "developed" image (in JPEG sRGB form, actually) that is to be examined.
Of course, for much of our work, we are concerned with the "dynamic range" potential of the camera as reflected in the raw output. Thus, even if we find the basic approach of the ISO definition "attractive", we have to adapt it to the raw data context. Among the wrinkles there is do we take the dynamic range, for example, to be the least of the dynamic ranges exhibited by the three color "channels", for testing with scene areas all of which have the chromaticity of some agreed standard illuminant, or what.
So while we may all be excited to learn that "the new Ulysses 99Z dSLR has a dynamic range of 42.1 dB (14 stops)", we need to be a little careful not to conclude that it will be "better" (with regard to the implications of dynamic range) than our Piccard D72, said to have a dynamic range of 37.6 dB (12.5 stops).
Now, it may well be that the camera industry has adopted some consistent basis on which to express dynamic range, and I just haven't heard about it (I'm an old guy, and I sleep a lot). And I'd be glad to learn of it.
Ah, my oatmeal is ready.
Tomorrow: Bran Chex
But when we begin comparing numerical expressions of the dynamic range of different cameras, we need to be mindful that there can be various ways to "score" that property, and not always will the same one have been used for the various "competitors".
The basic concept of course is straightforward: we consider the dynamic range of a camera (or imaging chain) to be the ratio of the largest and smallest luminance that can exist in a scene such that (in a single exposure) all the detail (of interest) of the scene is "usably captured".
It is the matter of "usably captured" that, in part, distinguishes the different possible quantitative definitions of dynamic range.
Note that when we say "luminance", we usually mean "base luminance" (my term), the "average" luminance of a scene area in which detail exists in the form of (small) variations in luminance above and below it.
One possible definition would be this:
The maximum luminance to be considered in the ratio is the highest one at which the slope of the exposure result curve (the digital equivalent of the "D log E" curve for film) is at least 90% of its value at a luminance a 1/2 stop lower.
The minimum luminance to be considered in the ratio is the highest one at which the slope of the exposure result curve is at least 90% of its value at a luminance 1/2 stop higher.
In effect, we consider the limits of the luminance range to be where "clipping" of detail has begun to emerge to a certain degree. (One could of course use different numerical criteria; my numbers were just to illustrate a type of definition.)
A comparable definition is one often used in connection with film.
But note that the value for "dynamic range" obtained under such a definition will greatly depend on the choice of those numbers in the definition.
However, some may argue that just because the imaging sensitivity retains its "incremental sensitivity" (a reasonable interpretation of the slope of the curve) at a certain small base luminance does not mean that we get a useful reproduction of detail about that luminance, because of the intrusion of noise.
Thus, we sometimes use a definition of dynamic range in which the "lower luminance" used in the ratio is the lowest luminance at which the "equivalent signal-to-noise ratio" is at least a certain stated target value (perhaps 1:1). (I say "equivalent" because of a subtlety that makes actual single-to-noise ratios as small as the usually-stated bogey impractical to measure directly.)
The ISO definition of the dynamic range of a digital camera is predicated on that concept for the lower luminance. In it, the "upper" luminance has a very simplistic definition: the luminance at which "sudden clipping" of the digital result occurs (as a result of exhausting the range of the numbering system used).
A further very important issue is where do we examine the "digital exposure result" to discern what scene luminances qualify as the limits of the luminance range (the numerator and denominator of our ratio). In the ISO definition, it is the "developed" image (in JPEG sRGB form, actually) that is to be examined.
Of course, for much of our work, we are concerned with the "dynamic range" potential of the camera as reflected in the raw output. Thus, even if we find the basic approach of the ISO definition "attractive", we have to adapt it to the raw data context. Among the wrinkles there is do we take the dynamic range, for example, to be the least of the dynamic ranges exhibited by the three color "channels", for testing with scene areas all of which have the chromaticity of some agreed standard illuminant, or what.
So while we may all be excited to learn that "the new Ulysses 99Z dSLR has a dynamic range of 42.1 dB (14 stops)", we need to be a little careful not to conclude that it will be "better" (with regard to the implications of dynamic range) than our Piccard D72, said to have a dynamic range of 37.6 dB (12.5 stops).
Now, it may well be that the camera industry has adopted some consistent basis on which to express dynamic range, and I just haven't heard about it (I'm an old guy, and I sleep a lot). And I'd be glad to learn of it.
Ah, my oatmeal is ready.
Tomorrow: Bran Chex