Doug Kerr
Well-known member
A recent thread in another section brought many discussions about white balance color correction. I thought I would reflect on some of the concepts.
I will talk about what I call theoretically-ideal white balance color correction. As I will mention shortly, this is not always what we want - maybe almost never. But (and I will repeat this later), But this is the only thing in this area for which we can have an objective definition, and thus can speak about "accuracy" and "correctness" and all those things beloved to us engineers.
I consider this to be the objective of theoretically-ideal white balance color correction (TIWBCC): The color recorded in an image file for a "patch" on the image corresponds to the reflective color of that patch of the subject. (I have omitted some details about the reference white and so forth to prevent obscuration of the principle.)
By way of a practical example, this means that in a shot of a letter (on "pure white" stationery) lying on a table, the recorded color of any place on the blank letter paper in the file will be neutral (for example, RGB=175,175,175).
Now this may well not be what we want. As Asher Kelman so clearly noted a while ago, if we have a shot of a cowboy reading a letter (on "pure white" stationery) by the yellow light of a campfire, we likely will want the paper recoded with a "yellowish" color in the image file - not an example of TIWBCC.
But, as I said earlier, TIWBCC is the only thing in this area for which we can have an objective definition, and thus can speak about "accuracy" and "correctness" and all those things beloved to us engineers.
By way of background, simplistically, the light reflected from our patch is influenced by both the reflective color of the patch and the chromaticity of the incident light.
The camera's sensor system responds to the color of the reflected from the patch; to determine the reflective color, we, in effect, have to (at some stage of the overall image processioning chain) "back out" the chromaticity of the incident light. And to do that, we must know the chromaticity of the incident light. We can get that in several ways:
a. We may already know (from a through specification of the light source, often available in studio work).
b. We can measure it with an incident light calorimeter placed at the subject location.
c. We can temporarily convert our camera to an incident light colorimeter by fitting it with a white balance measurement diffuser, and place it at the subject location.
d. We can, prior to the actual shot, place a neutral target (such as a "gray card) in the scene at the subject location and have the camera regard it (as part of a process which works different ways in different manufacturers' cameras of having the camera determine the presumed chromaticity of the incident light).
e. We can place in the scene for the actual shot, near the subject, a neutral target, and then regard the recorded color of that during some stage of external image processing (perhaps during raw development) and from that infer the chromaticity of the incident light.
f. We can draw upon some "natural" object in the scene whose reflective color is essentially neutral, using it as in method e.
If our plan if to have the camera perform the white balance color correction internally, while developing an image output (JPEG, TIFF, etc.), then we must use method a, b, c, or d. Methods c and d are most convenient, since they automatically "enter into the camera" the determined chromaticity of the incident light. (The camera rarely lets us enter into it directly a chromaticity in the form we will usually have it under method a or b.)
If our plan if to perform the white balance color correction while developing the raw file into an image output (JPEG, TIFF, etc.) outside the camera, then we must probably use method a, b, e, or f. Methods e-f are most convenient, since they automatically "enter into the software" the determined chromaticity of the incident light. (The software rarely lets us enter into it directly a chromaticity in the form we will usually have it under method a or b.)
Some cameras will report, in the metadata for subsequent images, the chromaticity value determined with method c or d, and there are ways to "sneak it out of the camera".
No color correction?
We sometimes hear the question, "suppose we do no white balance color correction". Well first, what might that mean?
Well, it would seem to mean that the camera would take the colors as observed by the camera's sensor chain and record them verbatim into the image file - period.
But for example, what chromaticity of light would result in the recording of a "neutral" color (such as RGB= 115,115,115)? Well, that would have to be the chromaticity of the white point of the output color space (sRGB, or may Adobe RGB).
Can we make the camera do that? Not usually exactly. The camera always does its development of the raw data in contemplation of some chromaticity of the ambient illumination.
Thus, to have "no white balance color correction", we must tell the camera that the chromaticity of the ambient illumination is the same as the white point chromaticity of the color space. And there is not usually a choice in the camera's menu of preset white balance corrections for that.
Best regards,
Doug
I will talk about what I call theoretically-ideal white balance color correction. As I will mention shortly, this is not always what we want - maybe almost never. But (and I will repeat this later), But this is the only thing in this area for which we can have an objective definition, and thus can speak about "accuracy" and "correctness" and all those things beloved to us engineers.
I consider this to be the objective of theoretically-ideal white balance color correction (TIWBCC): The color recorded in an image file for a "patch" on the image corresponds to the reflective color of that patch of the subject. (I have omitted some details about the reference white and so forth to prevent obscuration of the principle.)
By way of a practical example, this means that in a shot of a letter (on "pure white" stationery) lying on a table, the recorded color of any place on the blank letter paper in the file will be neutral (for example, RGB=175,175,175).
Now this may well not be what we want. As Asher Kelman so clearly noted a while ago, if we have a shot of a cowboy reading a letter (on "pure white" stationery) by the yellow light of a campfire, we likely will want the paper recoded with a "yellowish" color in the image file - not an example of TIWBCC.
But, as I said earlier, TIWBCC is the only thing in this area for which we can have an objective definition, and thus can speak about "accuracy" and "correctness" and all those things beloved to us engineers.
By way of background, simplistically, the light reflected from our patch is influenced by both the reflective color of the patch and the chromaticity of the incident light.
In fact, it is more complicated than this, and the real process involves both the reflective spectrum of the patch and the spectrum of the incident light. However, if both are "well behaved", we can work on the chromaticity basis I mentioned, and I will so assume here.
The camera's sensor system responds to the color of the reflected from the patch; to determine the reflective color, we, in effect, have to (at some stage of the overall image processioning chain) "back out" the chromaticity of the incident light. And to do that, we must know the chromaticity of the incident light. We can get that in several ways:
a. We may already know (from a through specification of the light source, often available in studio work).
b. We can measure it with an incident light calorimeter placed at the subject location.
c. We can temporarily convert our camera to an incident light colorimeter by fitting it with a white balance measurement diffuser, and place it at the subject location.
d. We can, prior to the actual shot, place a neutral target (such as a "gray card) in the scene at the subject location and have the camera regard it (as part of a process which works different ways in different manufacturers' cameras of having the camera determine the presumed chromaticity of the incident light).
e. We can place in the scene for the actual shot, near the subject, a neutral target, and then regard the recorded color of that during some stage of external image processing (perhaps during raw development) and from that infer the chromaticity of the incident light.
f. We can draw upon some "natural" object in the scene whose reflective color is essentially neutral, using it as in method e.
If our plan if to have the camera perform the white balance color correction internally, while developing an image output (JPEG, TIFF, etc.), then we must use method a, b, c, or d. Methods c and d are most convenient, since they automatically "enter into the camera" the determined chromaticity of the incident light. (The camera rarely lets us enter into it directly a chromaticity in the form we will usually have it under method a or b.)
If our plan if to perform the white balance color correction while developing the raw file into an image output (JPEG, TIFF, etc.) outside the camera, then we must probably use method a, b, e, or f. Methods e-f are most convenient, since they automatically "enter into the software" the determined chromaticity of the incident light. (The software rarely lets us enter into it directly a chromaticity in the form we will usually have it under method a or b.)
Some cameras will report, in the metadata for subsequent images, the chromaticity value determined with method c or d, and there are ways to "sneak it out of the camera".
No color correction?
We sometimes hear the question, "suppose we do no white balance color correction". Well first, what might that mean?
Well, it would seem to mean that the camera would take the colors as observed by the camera's sensor chain and record them verbatim into the image file - period.
But for example, what chromaticity of light would result in the recording of a "neutral" color (such as RGB= 115,115,115)? Well, that would have to be the chromaticity of the white point of the output color space (sRGB, or may Adobe RGB).
Can we make the camera do that? Not usually exactly. The camera always does its development of the raw data in contemplation of some chromaticity of the ambient illumination.
In fact, the application of a "vector" describing the color correction to be applied is usually integrated into the application of a vector taking into account the respective sensitivities of the three (or four) channels of a CFA sensor, and that process cannot be omitted.
Thus, to have "no white balance color correction", we must tell the camera that the chromaticity of the ambient illumination is the same as the white point chromaticity of the color space. And there is not usually a choice in the camera's menu of preset white balance corrections for that.
Best regards,
Doug