9 Replies Latest reply on Mar 2, 2015 9:03 AM by G.Hoffmann

    3200k color correction


      I would like to color correct some graphics to 3200k to save me from changing the settings on multiple monitors, do you know the color curve or the input details I would need to do this?

        • 1. Re: 3200k color correction
          G.Hoffmann Level 3

          Your question can be interpreted by two possibilities (examples):

          a) Simulate a scene which was shot under 5000K so, that it looks like shot under 3200K.

          b) Adjust an image, which looks correct on a 6500K monitor, so, that it looks alike on a

             3500K monitor (not common...).

          Please explain.


          In Adobe camera raw (ACR) the color temperature of a RAW image can be adjusted directly,

          using a Kelvin scale.


          Any other image (not RAW) can be brought into ACR, but only by going through Bridge.

          Normally, metadata are missing then. Now the color temperature can be adjusted only by

          using a relative scale -100 to +100.


          Make color and tonal adjustments in Camera Raw


          Furtheron, Photo Filter can be used, which is also not calibrated.


          Generally spoken, a color temperature adjustment requires a matrix multiplication of RGB values,

          which is in Photoshop not available, as far as I know. 


          Best regard  --Gernot Hoffmann

          • 2. Re: 3200k color correction

            Hi Gernot,


            could you explain why a color temperature adjustment requires a matrix multiplication.

            And what do you mean when you say photo filter is not calibrated?

            And one last question: when I change the color temperature in ACR, is it done with a matrix multiplication?


            Thanks in advance!

            • 3. Re: 3200k color correction
              G.Hoffmann Level 3

              Hello Jon,


              If two RGB-systems have different white-points, then the RGB values are related to each other

              by a matrix multiplication. If the two RGB systems have the same primaries, then the matrix is diagonal,

              which results in relations like this (with simplified numbers):

              Rd50 = 1.2 Rd65

              Gd50 = 1.0 Gd65

              Bd50 = 0.8 Bd65

              More red in D50, less blue.


              I'm not sure whether we really should dive into the mathematics.


              Nevertheless some references:

              I'm using a simple model.


              p. 10+14


              For more general cases (different primaries) or more accurate methods, a full matrix is required, but always

              with dominating diagonal values.

              Henry R.Kang

              Computational Color Technology

              SPI, 2006

              Chapter White-Point Conversion



              The white point correction in ACR is 'calibrated' (only for RAW sources), because it

              is related directly to source and target white-points, expressed by color temperatures

              (should be called correlated color temperatures):


              Photoshop Help Text:


              "Temperature: Sets the white balance to a custom color temperature. Decrease Temperature to correct a photo taken

              with a lower colortemperature of light; the Camera Raw plug-in makes the image colors bluer to compensate for the

              lower color temperature (yellowish) of theambient light. Conversely, increase Temperature to correct a photo taken with

              a higher color temperature of light; the image colors become warmer(yellowish) to compensate for the higher color

              temperature (bluish) of the ambient light.


              Moving the Temperature slider to the left corrects a photo taken with a lower color temperature of light."


              This is probably confusing, but in many cases these corrections are not applied by numbers, but

              by appearance (so-called 'preferred reproduction').

              In this sense can we understand Photoshops Photo-Filters. These reproduce in my understanding

              standardized optical filters but without referring directly to color temperature, therefore they are not



              I don't know the algorithms in ACR, but these white-point shifts are fairly good described scientifically

              as matrix operations, e.g. by H.R.Kang.


              Describing your actual task would help much...


              Best regards --Gernot Hoffmann

              • 4. Re: 3200k color correction
                fresh-prince Level 1

                Hello Gernot,


                thank you for your detailed answer. One thing I am not quite sure of is the following:


                You wrote that you think Photoshop isn't capable of doing matrix multiplications. I think what you mean is that Photoshop isn't capable of doing it, as a "standalone"-operation, right? Because when transforming between two rgb icc-profiles a matrix multiplication must be applied, or I am wrong? I am just curious


                And what about Color LookUp adjustment layers? Could you achieve a proper color temperature adjustment with them? (of course this probably requires look up tables ...)


                One other thing: beside raw-files photoshop cannot  tell you which color temperature one image currently has, or is there a way to "measure" that in Photoshop?


                Describing your actual task would help much...


                Sorry, but I am not the original poster, so I can't answer that question. I am just interested in this topic.


                Thank you very much!

                • 5. Re: 3200k color correction
                  G.Hoffmann Level 3

                  Yes, matrix operations cannot be applied by Photoshop users, perhaps by scripting.


                  Look-Up tables are not feasible because calculations are simple for programmers,

                  whereas tables would be huge – a True Color image (8 bits per channel) can contain 16.7

                  millions of colors.


                  Often one can restrict the task to shift the whitepoint using the same primaries. In such

                  a case one whitepoint is given, for instance D50 or D65, or in a digital foto "as shot" and

                  the other whitepoint is arbitrary. As already explained, the matrix between the two sets

                  RGB1 and RGB2 is diagonal.


                  For instance between D50 and D65 for Rec.709 primaries (sRGB) we have:


                  R1 = 1.1765 R2

                  G1 = 0.9756 G2

                  B1 = 0.7218 B2


                  R2 = 0.8500 R1

                  G2 = 1.0250 G1

                  B2 = 1.3855 B1


                  As an example a scene is given in sRGB (D65). The appearance of the scene under D50

                  should be simulated. Then we use the first set, where the image becomes warmer.

                  Images on p.14:



                  Fortunately it's possible to do this transform in Photoshop – by Curves. Just shift

                  the end points so, that curve Red becomes steeper, slope 1.1765, and for Green and Blue

                  less steep, slopes 0.9756 and 0.7128. 


                  Measuring the color temperature in an image requires the knowledge of an area which can

                  be considered as neutral (clouds, white eye, paper in a photo of a book...) For a human this

                  is not difficult, but for an automatic system? 


                  The RGB- or Lab-values can be measured. The chromaticities x and y are calculated.

                  Then the so-called correlated color temperature can be found:



                  Thanks for the interesting discussion.


                  Best regards --Gernot Hoffmann

                  • 6. Re: 3200k color correction
                    fresh-prince Level 1



                    I am already a lot smarter but I still have one question:


                    I tried to alter your example picture with a curve and the values given. According to my logic the middle picture "D65" (on which the curve is applied) and the left picture "D50" should look the same, or at least almost the same. But they don't.

                    In my tests it looks like the effect is too strong ... Could you explain what's going on here?

                    Here is a screenshot of what I did:



                    Thank you once again!

                    • 7. Re: 3200k color correction
                      G.Hoffmann Level 3

                      Thanks for your test. There is indeed something wrong.The effect is too strong.

                      By eyeball correction it should look like this:


                      I've to think about ...


                      Best regards --Gernot Hoffmann


                      I think I've got it:

                      The correction has to be applied in the linear light space:

                      Apply gamma encoding (parabola-like).

                      Apply color correction by curves.

                      Apply inverse gamma encoding (squareroot-like).

                      In the moment I don't know how to do this accurately by Photoshop.

                      My examples were calculated in Borland Pascal.

                      • 8. Re: 3200k color correction
                        fresh-prince Level 1

                        Hi Gernot,


                        I think I've got it:

                        The correction has to be applied in the linear light space:

                        Apply gamma encoding (parabola-like).

                        Apply color correction by curves.

                        Apply inverse gamma encoding (squareroot-like).

                        In the moment I don't know how to do this accurately by Photoshop.

                        My examples were calculated in Borland Pascal.


                        I tried that by converting the test image to 32 bit, which is always in linear light.

                        Unfortunately you can't use curves, but levels work fine for that. The result looks a lot better now.

                        I cannot really prove, if this is completely accurate but it seems to be quite close. What do you think about it?

                        Bildschirmfoto 2015-03-02 um 15.36.32.png


                        Anyway ... It seems like you've got it

                        Thanks for your inputs!

                        • 9. Re: 3200k color correction
                          G.Hoffmann Level 3

                          Thank you very much for the test. I didn't know that 32bpc works in linear light space.

                          The result is quite convincing. There may be small deviations if the screenshot

                          were not corrected by

                          - Assign monitor profile

                          - Convert to sRGB

                          - Save for Web

                          but this is not relevant here.


                          What is the meaning of this color temperature conversion?

                          The transformation calculates the measurable colorimetric data if the original scene

                          was shot under D65 and the modified scene would have been taken under D50.

                          There is nowhere any chromatic adaptation transform, because measuring devices

                          don't adapt, unless programmed for this purpose.


                          Photoshop workspace transformations from any workspace (e.g. one with D65 white

                          point ) to any other (e.g. one with D50), are using the so called Bradford Chromatic

                          Adaptation Transform (CAT).


                          In this case one has to deal with the appearance (opposed to measurable quantities).

                          In my old program under Windows 98 I can apply the color temperature conversion

                          as well by means of Bradford CAT. The result is only a little different to our old and

                          new example, but I can't reproduce it here, because the old PC doesn't accept my

                          USB stick...

                          I cannot exclude, that the old and new example is the Bradford version.


                          I would appreciate a private E-Mail about your identity. Should we know each other?


                          Best regards --Gernot Hoffmann