Doug Kerr
Well-known member
From time to time we have waves of interest in the specific performance of camera automatic exposure systems. Today I took some simplistic tests intended to investigate that matter for my EOS 40D.
BACKGROUND
Automatic exposure
Simplistically, the automatic exposure control system on most cameras enacts (or recommends) a camera exposure (shutter speed plus aperture) such that a metered exposure of a uniform-luminance scene will lead to a specific photometric exposure (illuminance-time product) on the film or digital sensor (for any given ISO sensitivity). Exactly what that photometric exposure is, for my EOS 40D, will be what I discuss here.
Under the ISO standards
By way of reference, if we have a camera that:
• has an automatic exposure system calibrated as recommended by the pertinent ISO standard
• has the ISO sensitivity rating (s) (an input to the automatic exposure calibration equation) that is determined on a "saturation" basic as defined by the pertinent ISO standard
then the resulting photometric exposure for the situation mentioned above should be 0.128 of the saturation exposure.
Now there are some complications in the concept of the saturation photometric exposure. For one thing, we cannot ordinarily actually determine that. We must instead infer it from the data delivered by the camera.
We might want to think of that in the context of the raw data delivered by the camera, or in terms of the RGB image delivered by the camera (in JPEG form), depending on the context in which we wish to make use of the information about exposure metering behavior..
In the case of the raw outlook, the notion is complicated by the differing sensitivities (and thus presumably differing photometric saturation levels) of the three "channels". In the case of the RGB (JPEG) outlook, it is further complicated by the application of white balance color correction and other image processing functions.
Returning to the matter of a the photometric exposure (for a uniform luminance subject) prescribed by the interaction of the two ISO standards, this can be interpreted in many ways. One popular interpretation is this (unfortunately, it is often mis-stated):
[We assume that the exposure metering uses a frame-wide average pattern, and otherwise follows the specifications above.]
If the camera regards a scene whose in-frame average reflectance is 18%, then any object in the scene with a reflectance of 100% will be given a photometric exposure that is 1/2 stop below saturation (sometimes said to be the "1/2 stop headroom premise).
MY 40D
Test setup
I tested by having my EOS 40D regard a neutral uniform luminance scene, essentially illuminated by mid-day open sun (North sky).
In fact the scene was the south end of a northbound white balance measurement diffuser.
The initial test exposure was taken with the evaluative pattern in effect (see a discussion of metering patterns later) and zero exposure compensation. Exposures were then taken with the same setup and exposure compensation of +1.0 Ev and -1.0 Ev.
Raw data evaluation
The raw files were loaded into Rawnalyze and the three-channel histogram evaluated. We used as an indicator of photometric exposure the upper edge of the histogram peak, in particular the 99.9 percentile edge (the level above which only fell 0.1% of the pixels).
Of course, that level was different for the three sensor channels, with the G channel highest. We used the G channel data as our indicator of "approach to saturation".
We reckoned the upper edge value against the saturation level (as reported by the camera in the exif metadata and displayed by Rawnalyze, considering the bottom edge of the range to be the black point as reported by Rawnalyze.
Before proceeding, I confirmed that the data (as interpreted as just described) was consistent between the 0 Ec, +1 Ec, and -1 Ec exposures. There was in fact good consistency between them - the change in exposure produced a linear change in the critical exposure result.
The result
For the test metered exposure, the exposure result suggested a photometric exposure for the highest-edge of the (narrow) distribution of the G channel histogram of 0.168 of saturation.
That same level corresponds to 0.237 of a "ceiling" that is 1/2 stop below saturation.
This is an exposure that is about 0.4 stop more aggressive that that implied by the ISO standards I referred to earlier.
This is generally consistent with what I have understood has been Canon's plan for EOS digital cameras for some while.
In the JPEG file
I developed the raw file in DPP. I set the "faithful" picture style and applied white balance color correction based on the image (which was of a "neutral" subject).
The result
The resulting RGB image had most pixel values in the area of 120,120,120. This implies a photometric exposure of 0.187 of that corresponding to "saturation" (in the sense of RGB values of 255).
METERING PATTERNS
There is no "correct" relationship between the calibrations of exposure metering with different patterns. Only the "frame-wide average" pattern really fits into the theoretical story I have told here (and we don't even have that available in our EOS cameras).
But it is interesting to compare the metered result for the different patterns when the camera regards a uniform-luminance scene.
Here, that result was consistent within 0.1 stop over the four available patterns: evaluative, center-weighted average, partial, and spot.
CANON'S PLAN
One might wonder why Canon does not use a combination of exposure metering calibration and ISO rating that produces an exposure about 1/2 stop hotter than would be expected under the recommendations of the ISO standards.
But I would never presume to know what Canon has in mind.
However, I would think that this is largely a result of the fact that the evaluative metering system is supposed to use its intelligence to produce a more optimal exposure than would basic frame-wide average metering. One result would be that less "headroom" would ordinarily be required for safety. (Remember, the ISO strategy is based on the wholly arbitrary "assumption" of a scene with an average reflectance of 18%. The headroom is largely needed for cases in which the actual scene reflectance distribution has a lower average.)
ACTUAL PRACTICE
I will not here attempt to say what guidance the practicing photographer should take from the data I report.
BACKGROUND
Automatic exposure
Simplistically, the automatic exposure control system on most cameras enacts (or recommends) a camera exposure (shutter speed plus aperture) such that a metered exposure of a uniform-luminance scene will lead to a specific photometric exposure (illuminance-time product) on the film or digital sensor (for any given ISO sensitivity). Exactly what that photometric exposure is, for my EOS 40D, will be what I discuss here.
Under the ISO standards
By way of reference, if we have a camera that:
• has an automatic exposure system calibrated as recommended by the pertinent ISO standard
• has the ISO sensitivity rating (s) (an input to the automatic exposure calibration equation) that is determined on a "saturation" basic as defined by the pertinent ISO standard
then the resulting photometric exposure for the situation mentioned above should be 0.128 of the saturation exposure.
Now there are some complications in the concept of the saturation photometric exposure. For one thing, we cannot ordinarily actually determine that. We must instead infer it from the data delivered by the camera.
We might want to think of that in the context of the raw data delivered by the camera, or in terms of the RGB image delivered by the camera (in JPEG form), depending on the context in which we wish to make use of the information about exposure metering behavior..
In the case of the raw outlook, the notion is complicated by the differing sensitivities (and thus presumably differing photometric saturation levels) of the three "channels". In the case of the RGB (JPEG) outlook, it is further complicated by the application of white balance color correction and other image processing functions.
Returning to the matter of a the photometric exposure (for a uniform luminance subject) prescribed by the interaction of the two ISO standards, this can be interpreted in many ways. One popular interpretation is this (unfortunately, it is often mis-stated):
[We assume that the exposure metering uses a frame-wide average pattern, and otherwise follows the specifications above.]
If the camera regards a scene whose in-frame average reflectance is 18%, then any object in the scene with a reflectance of 100% will be given a photometric exposure that is 1/2 stop below saturation (sometimes said to be the "1/2 stop headroom premise).
MY 40D
Test setup
I tested by having my EOS 40D regard a neutral uniform luminance scene, essentially illuminated by mid-day open sun (North sky).
In fact the scene was the south end of a northbound white balance measurement diffuser.
The initial test exposure was taken with the evaluative pattern in effect (see a discussion of metering patterns later) and zero exposure compensation. Exposures were then taken with the same setup and exposure compensation of +1.0 Ev and -1.0 Ev.
Raw data evaluation
The raw files were loaded into Rawnalyze and the three-channel histogram evaluated. We used as an indicator of photometric exposure the upper edge of the histogram peak, in particular the 99.9 percentile edge (the level above which only fell 0.1% of the pixels).
Of course, that level was different for the three sensor channels, with the G channel highest. We used the G channel data as our indicator of "approach to saturation".
We reckoned the upper edge value against the saturation level (as reported by the camera in the exif metadata and displayed by Rawnalyze, considering the bottom edge of the range to be the black point as reported by Rawnalyze.
Before proceeding, I confirmed that the data (as interpreted as just described) was consistent between the 0 Ec, +1 Ec, and -1 Ec exposures. There was in fact good consistency between them - the change in exposure produced a linear change in the critical exposure result.
The result
For the test metered exposure, the exposure result suggested a photometric exposure for the highest-edge of the (narrow) distribution of the G channel histogram of 0.168 of saturation.
That same level corresponds to 0.237 of a "ceiling" that is 1/2 stop below saturation.
This is an exposure that is about 0.4 stop more aggressive that that implied by the ISO standards I referred to earlier.
This is generally consistent with what I have understood has been Canon's plan for EOS digital cameras for some while.
In the JPEG file
I developed the raw file in DPP. I set the "faithful" picture style and applied white balance color correction based on the image (which was of a "neutral" subject).
The result
The resulting RGB image had most pixel values in the area of 120,120,120. This implies a photometric exposure of 0.187 of that corresponding to "saturation" (in the sense of RGB values of 255).
METERING PATTERNS
There is no "correct" relationship between the calibrations of exposure metering with different patterns. Only the "frame-wide average" pattern really fits into the theoretical story I have told here (and we don't even have that available in our EOS cameras).
But it is interesting to compare the metered result for the different patterns when the camera regards a uniform-luminance scene.
Here, that result was consistent within 0.1 stop over the four available patterns: evaluative, center-weighted average, partial, and spot.
CANON'S PLAN
One might wonder why Canon does not use a combination of exposure metering calibration and ISO rating that produces an exposure about 1/2 stop hotter than would be expected under the recommendations of the ISO standards.
But I would never presume to know what Canon has in mind.
However, I would think that this is largely a result of the fact that the evaluative metering system is supposed to use its intelligence to produce a more optimal exposure than would basic frame-wide average metering. One result would be that less "headroom" would ordinarily be required for safety. (Remember, the ISO strategy is based on the wholly arbitrary "assumption" of a scene with an average reflectance of 18%. The headroom is largely needed for cases in which the actual scene reflectance distribution has a lower average.)
ACTUAL PRACTICE
I will not here attempt to say what guidance the practicing photographer should take from the data I report.