1 2 Previous Next 48 Replies Latest reply on Feb 24, 2009 12:58 PM by Tim Lookingbill

    Exposure slider not entirely linear ?

    Peter DL Level 1
      Ive recently captured an image with a bright, red to orange sky in the background due to sunset conditions. With a linear preset in ACR (all tonal controls zero, parametric and point curve linear) theres no clipping indicated.

      Actually the Exposure slider has to be raised to +0.70 in order to drive relevant parts of the sky into clipping. But, the Exposure slider has to be set down to -0.70 in order to detach the histogram from the right end of scale.

      In between, its obvious that the histogram does not react linearly to the Exposure slider. Increasing it from 0.70 to +0.70 accumulates and compresses more and more data in the right corner of the histogram (up to RGB 254) without making the final step to clipping.

      Output space is set to ProPhoto RGB, Noise reduction and Sharpening are off, the Recovery slider for sure is at zero. Its CS4 + ACR 5.2.

      Any thoughts or explanations?
      Thanks!

      Peter

      --
        • 1. Re: Exposure slider not entirely linear ?
          Panoholic Level 2
          The "exposure adjustment" occurs
          b before
          the non-linear transformation occurs (which is dictated by the color space selection). Thus the linear adjustment too becomes transformed by the respective power function.

          The effect can be observed better on gray scales; pick some more or less gray point with the color sampler in the very highlights, one in the mid shades and another in the very shadows. Change now the "exposure" and observe the changes in the RGB values of the two sampled points: the low RGB values changes much faster than the high ones.

          Furthermore, the range to the right edge is "open ended": if you push it more and more, it can't go further to the right in the calculated pixel value range and in the displayed histogram. When you reduce the "exposure", it may change in the clipped region without moving to the "inside of the range".
          • 2. Re: Exposure slider not entirely linear ?
            Peter DL Level 1
            > The "exposure adjustment" occurs before the non-linear transformation occurs (which is dictated by the color space selection). Thus the linear adjustment too becomes transformed by the respective power function.

            Did a quick test as follows: The image was processed three times with different Exposure settings in ACR. In Photoshop the three files were converted from ProPhoto RGB to a linear gamma version of ProPhoto RGB, then overlaid and a measuring point was set to the sky.

            Exposure +0.6: RGB= 252, 253, 186
            Exposure 0.0: RGB= 247, 249, 140
            Exposure 0.7: RGB= 239, 187, 110

            The scaling factor with the Red channel is lower compared to the Blue one (252/239 < 186/110). The speed of the Green channel seems to be in-between.

            Ill have to double-check tomorrow.
            For this evening I find this puzzling.

            Many thanks for your comment.
            Peter

            --
            • 3. Re: Exposure slider not entirely linear ?
              boscarol Level 1
              I did this experiment.

              In ACR, I set all slides to default values and Blacks to 0.
              Then I set the output space to ProPhoto.
              Then I choose a linear DNG profile for Nikon D40 (my camera) in Camera Calibration.

              Finally, I devise a method to translate from RGB readouts to XYZ (normalization and multiplication for the ProPhoto->XYZ matrix).

              Now, the results for a selected pixel (RGB = 30 30 32) are:

              -Exposure 0: XYZ = 0.0206 0.0212 0.0197
              -Exposure 1: XYZ = 0.0401 0.0415 0.0371 (a factor 2 expected, but a factor 1.95 obtained)
              -Exposure 2: XYZ = 0.0795 0.0836 0.0734 (a factor 4 expected, but a factor 3.87 obtained)

              I expected factor 1, 2 and 4, but obtained factor 1, 1.95, 3.87. Same factor for X, Y and Z (and this is a good thing) but not 1, 2, 4.

              Can someone comment for this (little) difference?

              Thank you.

              Mauro Boscarol
              • 4. Re: Exposure slider not entirely linear ?
                Panoholic Level 2
                > In Photoshop the three files were converted from ProPhoto RGB to a linear gamma version of ProPhoto RGB

                This is not an argument against or for anything, but how did you do that? I mean, where did you get a ProPhoto profile with a "linear gamma"? (I always shudder when using the term "linear gamma", but I am forced to.)

                > The scaling factor with the Red channel is lower compared to the Blue one (252/239 < 186/110). The speed of the Green channel seems to be in-between

                That's the reason I suggested to pick gray points. The transformation required to maintain the saturation is a fixed in the color space definitions. You can trust only gray levels: they are supposed to remain gray at whatever intensity level. However, the proportions of the R, G and B components
                b can not
                stay fixed in the entire range. Think of a color expressed by RGB=(1,50,100). Now, how about increasing the intensity to R=3? What about to increase R to 100?
                • 5. Re: Exposure slider not entirely linear ?
                  Jao vdL Adobe Community Professional & MVP
                  >This is not an argument against or for anything, but how did you do that? I mean, where did you get a ProPhoto profile with a "linear gamma"?

                  No need for a custom profile, that's fairly simple to do with the convert to profile dialog and by hitting "custom RGB". Select gamma 1, D50 whitepoint and for the red, green, and blue primaries the following x and y's:

                  red 0.7347 0.2653
                  green 0.1596 0.8404
                  blue 0.0366 0.0001
                  • 6. Re: Exposure slider not entirely linear ?
                    Panoholic Level 2
                    > No need for a custom profile, that's fairly simple to do with the convert to profile dialog and by hitting "custom RGB"

                    Thanks; I read this already, even checked it out but never used and forgotten. I'll ask this again in a short while :-)
                    • 7. Re: Exposure slider not entirely linear ?
                      Peter DL Level 1
                      > That's the reason I suggested to pick gray points. The transformation required to maintain the saturation is a fixed in the color space definitions. You can trust only gray levels: they are supposed to remain gray at whatever intensity level. However, the proportions of the R, G and B components can not stay fixed in the entire range. Think of a color expressed by RGB=(1,50,100). Now, how about increasing the intensity to R=3? What about to increase R to 100?

                      My understanding was that linear scaling applies to every single pixel and every single R,G, B number per pixel. For illustration take a RGB composite curve i.e. a straight line with a slope > 1 (in Photoshop). All RGB triplets are multiplied according to:
                      (RGB)2 = f x (RGB)1
                      2 and 1 are meant as indices.
                      Hence its irrelevant if this a gray patch or a color.

                      Interestingly HSB hue and saturation are maintained which corresponds to:
                      (RGB)2 = f x (RGB)1
                      R2 : G2 : B2 = (f x R1) : (f x G1) : (f x B1)
                      R2 : G2 : B2 = R1 : G1 : B1
                      Means that the intensity ratios (per RGB triplet) do not change.

                      Clipping is of course a limitation. When you elevate R= 1 to 3 by applying a respective factor, G changes from 50 to 150, whereas B= 100 is driven into clipping of 255. However, clipping was barely avoided with my test described above.

                      In the meantime it turned out that the camera profile is of some influence. I was using a custom profile created via the DNG profile editor. However, changing to the baseline matrix profile it still doesnt show me the expected linearity of the Expose slider with this sunset sky. At least not according to the numbers with my testing procedure...

                      Peter

                      --
                      • 8. Re: Exposure slider not entirely linear ?
                        Panoholic Level 2
                        > My understanding was that linear scaling applies to every single pixel and every single R,G, B number per pixel

                        You started out with ACR, not with Photoshop. The conversion of the demosaiced raw data from the camera's color space in sRGB, aRGB or ProPhoto RGB includes the non-linear transformation as well. This can not be avoided in ACR to my knowledge; Canon's DPP does support linear output (I would be happy to hear, that ACR too can do this).

                        For example if you convert the linear data to sRGB, following function will be applied:

                        f(x) = 12.92x if x <= 0.0031308
                        f(x) = 1.055 * (x ^ (1/2.4)) - 0.055 if x > 0.0031308

                        This transformation is reversible mathematically, but not reversible in digital photography with integer values for pixels; one can regard this transformation as a lossy compression (in fact the lossy compression adopted in many Nikon cameras closely follows this function). Thus there is not much sense to linearize the image data after it had been transformed already.
                        • 9. Re: Exposure slider not entirely linear ?
                          MadManChan2000 Adobe Employee
                          You can effectively get linear output from Camera Raw (with the exception of gamma encoding) by creating a custom color profile with the DNG Profile Editor with its Base Tone Curve set to Linear, then use the Camera Raw Defaults when processing your image and select that custom color profile. As Gabor notes, the result will be gamma-encoded using your choice of RGB color space. Having linear output may be useful if you're trying to feed the image to another software program that works best on linear data (e.g., some tone mapping software).
                          • 10. Re: Exposure slider not entirely linear ?
                            boscarol Level 1
                            Eric,

                            I use a custom linear DNG camera profile and I calculate XYZ for a pixel.

                            I get a XYZ value for exposure set to 0, and XYZ * 1.95 for exposure set to 1.

                            I expected XYZ *2. (See my message in this thread above for more details.)

                            Can you help me find the possible cause of this?

                            Thank you.

                            Mauro Boscarol
                            • 11. Re: Exposure slider not entirely linear ?
                              Bill_Janes Level 2
                              >You can effectively get linear output from Camera Raw (with the exception of gamma encoding) by creating a custom color profile with the DNG Profile Editor with its Base Tone Curve set to Linear, then use the Camera Raw Defaults when processing your image and select that custom color profile

                              Eric, can't you also get linear results (except for gamma encoding) by simply setting all the values on the basic ACR panel to zero and also setting the point tone curve to zero?

                              For example, I investigated Peter's question of linearity of the ACR exposure control by shooting a Stouffer step wedge and then rendering into aRGB. I then set the exposure slider to -1 EV and repeated the process.

                              When plotted log-log as for a standard characteristic curve, I found the exposure compensation to be linear over a DR of 1000 (10 f/stops) as shown in the linked graphic.

                              http://bjanes.smugmug.com/photos/467657250_EJQQC-O.gif
                              • 12. Re: Exposure slider not entirely linear ?
                                Bill_Janes Level 2
                                >You can effectively get linear output from Camera Raw (with the exception of gamma encoding) by creating a custom color profile with the DNG Profile Editor with its Base Tone Curve set to Linear, then use the Camera Raw Defaults when processing your image and select that custom color profile

                                Eric, can't you also get linear results (except for gamma encoding) by simply setting all the values on the basic ACR panel to zero and also setting the point tone curve to zero?

                                For example, I investigated Peter's question of linearity of the ACR exposure control by shooting a Stouffer step wedge and then rendering into aRGB. I then set the exposure slider to -1 EV and repeated the process.

                                When plotted log-log as for a standard characteristic curve, I found the exposure compensation to be linear over a DR of 1000 (10 f/stops) as shown in the linked graphic.

                                http://bjanes.smugmug.com/photos/467657250_EJQQC-O.gif
                                • 13. Re: Exposure slider not entirely linear ?
                                  Peter DL Level 1
                                  > You can effectively get linear output from Camera Raw (with the exception of gamma encoding) by creating a custom color profile with the DNG Profile Editor with its Base Tone Curve set to Linear, then use the Camera Raw Defaults when processing your image and select that custom color profile.

                                  Eric, - Many thanks for joining discussion. So what would be the difference between a.) using such linearized custom profile created the way you describe with the DNG profile editor, and b.) employing a linear preset in ACR, means to set all tonal controls to zero, parametric/point curve linear while using the baseline matrix profile ?

                                  Or, let me ask straightforward: could there possibly be something behind my question, or am I barking up the wrong tree ?


                                  > The conversion of the demosaiced raw data from the camera's color space in sRGB, aRGB or ProPhoto RGB includes the non-linear transformation as well. This can not be avoided in ACR to my knowledge

                                  Again my understanding was that matrix-to-matrix conversion is essentially a linear transform (if this is the right term). Means that linear scaling and matrix-to-matrix conversion are essentially commutative, the sequence can be exchanged and the results will be the same (provided that the absolute value of the scaling factor is adjusted) and despite different matrix primaries and different gamma. However, I agree that there are a couple of constraints such as:
                                  /> out of gamut colors
                                  /> fancy 'gamma' e.g. with the sRGB TRC
                                  /> AbsCol conversion between different white points which can introduce additional clipping
                                  /> RelCol conversion between different white points (not sure about this one)

                                  So in order to check my approach of testing, I did some ColorChecker exercises. Processed it through ACR (linear preset) with different Exposure settings. Converted to 'linear gamma' ProPhoto RGB in Photoshop. Finally, all derived ratios R2:R1, G2:G1 and B2:B1 were approx the same, whether this is a color or gray patch which I think confirms linearity.

                                  Seems I still have to find out whats so special with my sunset sky. Or, the hue preserving feature comes in mind which was reported to cover ACRs tone curve(s). Could it be that it acts 'on' the Exposure slider as well, once a single channel is driven into clipping, thus, preventing this. Just a thought.


                                  Many thanks for your comments.
                                  Peter

                                  --
                                  • 14. Re: Exposure slider not entirely linear ?
                                    boscarol Level 1
                                    >Eric, can't you also get linear results (except for gamma encoding) by simply setting all the values on the basic ACR panel to zero and also setting the point tone curve to zero?

                                    No. See this thread

                                    http://www.adobeforums.com/webx?128@@.59b71da3/0

                                    Mauro
                                    • 15. Re: Exposure slider not entirely linear ?
                                      Panoholic Level 2
                                      > my understanding was that matrix-to-matrix conversion is essentially a linear transform

                                      Yes, but

                                      1. the color space transformation depends on the white point,

                                      2. a non-linear transformation (the "gamma encoding") too will be performed.

                                      If you convert the raw data in 16bit RGB TIFF in ACR, then convert it to "gamma = 1" in PS, the gamma encoding becomes reversed - but what about the white point? When defining the "linear RGB" in PS, you would have to supply the very same values, which were used by ACR or the DNG converter.

                                      Note, that I am not sure of the soundness of this line of thought. I tried to verify this by tests; the reverse linearization worked more or less, but never exactly, and the dark and bright ends were far from being linear. I did this with different white points in the Custom RGB spec, and the resulting proportions changed.

                                      > Seems I still have to find out whats so special with my sunset sky

                                      You have yet to state your actual problem. You started out with the unexpected result from the "exposure" adjustment of ACR, then switched to the linearization in Photoshop.

                                      So, what is so special with your sunset sky, beside not reacting to exposure adjustment as rapidly as you are expecting it?
                                      • 16. Re: Exposure slider not entirely linear ?
                                        Bill_Janes Level 2
                                        Bill Janes wrote:
                                        >Eric, can't you also get linear results (except for gamma encoding) by simply setting all the values on the basic ACR panel to zero and also setting the point tone curve to zero?

                                        Mauro Boscarol wrote:
                                        >No. See this thread

                                        The method I mentioned may not be the best one to obtain a linear file, but that method is often quoted and it does produce a linear result.

                                        To prove this, I used a shot of a Stouffer stepwedge and rendered it into 16 bit ProPhotoRGB with the ACR "linear" settings and then converted it to ProPhotoRGB_Gamma1 (a custom profile that I created in Photoshop) and plotted resulting curves with Imatest. I then combined the two curves in Excel.

                                        As is evident the "linear curve" is encoded with a gamma of 1/1.8 and the curve after converting to ProPhotoRGB_Gamma1 is very nearly linear over a range of 10 f/stops

                                        ProPhotoRGB result:
                                        http://bjanes.smugmug.com/photos/468061029_ZdjoD-O.png

                                        ProPhotoRGB linear result:
                                        http://bjanes.smugmug.com/photos/468061022_hVVXN-O.png

                                        Excel plot of both curves:
                                        http://bjanes.smugmug.com/photos/468057381_AhJLL-O.gif

                                        Perhaps Eric or Peter can comment.
                                        • 17. Re: Exposure slider not entirely linear ?
                                          Peter DL Level 1
                                          > No. See this thread
                                          http://www.adobeforums.com/webx?128@@.59b71da3/0

                                          Many thanks for sharing the link. If I understand correctly, the medium contrast point curve plus Brightness 50 & Contrast 25 together with a linear profile from DNG PE is recommended as the method of choice to get linear output from Camera Raw (with the exception of gamma encoding).

                                          Still not sure about the difference between both options a.) [Matrix profile w/linear curve + Brightness 50 & Contrast 25 + medium contrast Point curve] and b) [Matrix profile w/unchanged base tone curve + Brightness 0 & Contrast 0 + linear Point curve].


                                          > Yes, but
                                          1. the color space transformation depends on the white point,
                                          2. a non-linear transformation (the "gamma encoding") too will be performed.

                                          Regarding different gamma, said commutativity should still work as far as regular exponents are concerned (see below equation). But, I would not exclude deviations, particularly in the shadows due to the slope limiting feature of the Adobe color conversion engine (q.e.d).
                                          (f x rgb)^(1/g) = f^(1/g) x rgb^(1/g)

                                          > You started out with the unexpected result from the "exposure" adjustment of ACR, then switched to the linearization in Photoshop.

                                          Thinking about it, we can skip the trip to Photoshop, means just to take the RGB readings in ACR which refer to the selected output space afaik. Taking ProPhoto RGB it meets the requirement to feature a regular 1.8 gamma. Some further tests:


                                          I.) Color sampler in ACR set to the yellow patch of a ColorChecker capture

                                          Linearization procedure a)

                                          Exposure 0.0: RGB= 164, 148, 59
                                          Exposure +0.75: RGB= 208, 189, 77
                                          Exposure +1.20: RGB= 247, 225, 92

                                          208 : 164 = 189 : 148 = 77 : 59 at least approximately
                                          247 : 164 = 225 : 148 = 92 : 59 at least approximately

                                          With linearization procedure b) all obtained RGB numbers are +/-1 level the same


                                          II.) Color sampler in ACR set to the sunset sky of initially mentioned image

                                          Linearization procedure a)

                                          Exposure -1.40: RGB= 206, 188, 136
                                          Exposure -0.65: RGB= 247, 212, 162

                                          247 : 206 = 212 : 188 = 162 : 136 at least approximately

                                          With linearization procedure b) all obtained RGB numbers are +/-1 level the same


                                          Peter

                                          --
                                          • 18. Re: Exposure slider not entirely linear ?
                                            MadManChan2000 Adobe Employee
                                            (Note: I will return later to comment on this thread; at the moment I'm a bit swamped.)
                                            • 19. Re: Exposure slider not entirely linear ?
                                              Level 1
                                              > (Note: I will return later to comment on this thread; at the moment I'm a bit swamped.)

                                              Because you're adding a bazillion cool new features to next release of LR? :-)
                                              • 20. Re: Exposure slider not entirely linear ?
                                                ssprengel Adobe Community Professional & MVP
                                                > Because you're adding a bazillion cool new features to next release of LR?

                                                Eric Chan is a member of the Camera RAW Engineering team.
                                                • 21. Re: Exposure slider not entirely linear ?
                                                  Level 1
                                                  Eric is on the Camera Raw team yes...but that includes the Camera Raw pipeline in Lightroom's Develop as well. What is done in one is in the other as well.
                                                  • 22. Re: Exposure slider not entirely linear ?
                                                    Level 1
                                                    I'm well aware of who Eric is. It was a joke in case anyone misinterprets.
                                                    • 23. Re: Exposure slider not entirely linear ?
                                                      MadManChan2000 Adobe Employee
                                                      Hi Dave, yes, my threads are quite full at the moment ... :-)
                                                      • 24. Re: Exposure slider not entirely linear ?
                                                        Peter DL Level 1
                                                        In the meantime, I did another test (see post # 17) now using a linear version of my Custom profile which was created via the Chart Wizard of the DNG Profile editor. Again both procedures for linearization: a.) and b.) give identical RGB data within 1 level deviation. Further, the idea: simply to use the RGB readings in ACR even though these are gamma encoded, and to analyze linearity by building ratios of R2 : R1, G2 : G1, B2 : B1 (for different Exposure settings 1 and 2), stems from above mentioned 'commutativity' of linear scaling i.e. multiplication and gamma encoding i.e. application of an exponent.

                                                        II.) continued: Linearization procedure a.) now using a linear version of my Custom profile

                                                        Exposure -1.20: RGB= 217, 198, 120
                                                        Exposure -0.75: RGB= 230, 221, 138
                                                        Exposure 0.0: RGB= 247, 250, 154

                                                        247 : 217 = 1.14 is not equal (even not what I would have called approximately)
                                                        to 225 : 148 = 1.26 and 154 : 120 = 1.28


                                                        Peter

                                                        --
                                                        • 25. Re: Exposure slider not entirely linear ?
                                                          MadManChan2000 Adobe Employee
                                                          > Eric, can't you also get linear results (except for gamma
                                                          > encoding) by simply setting all the values on the basic
                                                          > ACR panel to zero and also setting the point tone curve to
                                                          > zero?

                                                          Yes, that works too and is equivalent to creating a "linear
                                                          base tone curve" DNG profile using the DNG Profile Editor.
                                                          • 26. Re: Exposure slider not entirely linear ?
                                                            MadManChan2000 Adobe Employee
                                                            Peter, in general a (linear) matrix op and the tone curve application will only be commutative if the tone curve is strictly a gamma curve (i.e., can be expressed as f(x) = x^g for some constant g). You will see some discrepancies in the shadows if using a modified version of a gamma curve with a limited slope (designed intentionally so that the derivative of the curve doesn't go to infinity at zero).
                                                            • 27. Re: Exposure slider not entirely linear ?
                                                              Peter DL Level 1
                                                              > Yes, that works too and is equivalent to creating a "linear
                                                              base tone curve" DNG profile using the DNG Profile Editor.

                                                              Eric, - Many thanks for clarification.

                                                              Some slight concerns come in my mind over the "Medium Contrast" Point Curve (which I think is some kind of Luminosity curve) and making it part of the linearized state together with a "linear base tone curve DNG profile". Normally such color model does perfectly maintain the R:G:B intensity ratios per color (afaik). HSB saturation drops with increasing Luminosity and vice versa. Anyway.

                                                              > Peter, in general a (linear) matrix op and the tone curve application will only be commutative if the tone curve is strictly a gamma curve (i.e., can be expressed as f(x) = x^g for some constant g). You will see some discrepancies in the shadows if using a modified version of a gamma curve with a limited slope (designed intentionally so that the derivative of the curve doesn't go to infinity at zero).

                                                              Right. There are limitations such as "fancy gamma", and in the deep shadows I assume the Slope limiting feature of the color conversion engine may even distort a regular gamma function enough to affect "commutativity" a bit. However, I was more focused on the highlights. While the output space was set to regular 1.8 gamma ProPhoto RGB, it seems to me still correct to work with the ACR RGB readings and to check them for a common multiplier "between" different Exposure settings. Please correct me if Im wrong.

                                                              Anyway, I have to admit that for now I cant reproduce the magnitude of the effect as originally seen. Oops. Upon more careful testing some non-linearity was found with my Custom profile (created with the Chart wizard), but thats probably not surprising, its the nature of such Lut type profile. Finally, its a very little price to pay (if any at all, depending on the point of view) for the much better profile and resulting color rendition as I find. That said, Im wondering what it means for ETTR though I dont practice it a lot.

                                                              Peter

                                                              --
                                                              • 28. Re: Exposure slider not entirely linear ?
                                                                Bill_Janes Level 2
                                                                Eric Chan wrote:

                                                                >...in general a (linear) matrix op and the tone curve application will only be commutative if the tone curve is strictly a gamma curve (i.e., can be expressed as f(x) = x^g for some constant g). You will see some discrepancies in the shadows if using a modified version of a gamma curve with a limited slope (designed intentionally so that the derivative of the curve doesn't go to infinity at zero).

                                                                Eric,

                                                                Perhaps you can clear up some confusion on my part. In reading the specs sRGB and ProPhotoRGB, I see that both have linear segments in the shadows. The aRGB 1998 spec specifically states that there is no linear segment in this space.

                                                                Then one sometimes reads about "simplified" sRGB as used by Adobe and I have regarded this as eliminating the linear segment and using a straight gamma 2.2 curve.

                                                                When one uses sRGB and ProPhotoRGB in PS and ACR is there a linear segment in the shadows according to spec or is a "simplified" tone curve used?
                                                                • 29. Re: Exposure slider not entirely linear ?
                                                                  Peter DL Level 1
                                                                  > When one uses sRGB and ProPhotoRGB in PS and ACR is there a linear segment in the shadows according to spec or is a "simplified" tone curve used?

                                                                  Bill, - its just me. The ProPhoto RGB space which ships with Photoshop features TRC tags in simple gamma style (rTRC, gTRC, bTRC). As I understand, its just a straight 1.8 gamma. Whereas with sRGB the TRC tags bear a 1024 points curve, respectively (see ICC profile inspector), as the case may be to describe the deviation from regular 2.2 gamma.

                                                                  For verification its worth to note that the Photoshop Custom profile function does not allow to tranfer such point curve TRC tags (afaik). So when you start with sRGB as shipped you get a Simplified sRGB with regular 2.2 gamma. The term "Simplified" is explicitely stated in the Custom profile function menu. After saving such profile it just requires to assign this or that profile to an image in order to see the differences. This is not so with ProPhoto RGB.

                                                                  In summary, Photoshops ProPhoto RGB space is obviously regular 1.8 gamma, whereas sRGB seems to follow the specs, thus deviating from regular 2.2 gamma in the shadows. Id be surprised if ACR including its RGB readings work differently, however, it certainly requires Eric for final confirmation.

                                                                  Peter

                                                                  --
                                                                  • 30. Re: Exposure slider not entirely linear ?
                                                                    MadManChan2000 Adobe Employee
                                                                    Bill, Peter, I cannot speak for PS (since I don't work on it directly) but in Camera Raw (and Lightroom), the color spaces sRGB, Adobe RGB, and ProPhoto RGB all use a non-gamma segment in the shadows. For developers that are interested in the details (or anybody who is familiar with C++), please see the DNG SDK, source file dng_color_space.h.
                                                                    • 31. Re: Exposure slider not entirely linear ?
                                                                      Bill_Janes Level 2
                                                                      Eric and Peter,

                                                                      It is interesting to learn that ACR uses a linear segment in the shadows even for aRGB, which normally does not have such a linear segment. Peter is quite knowledgeable, and I assume that Photoshop behaves as he states.

                                                                      One might ask what is the practical significance of a linear segment in the shadows? I suspect that the significance is minimal since the shadows are clipped significantly in normal photograph use.

                                                                      These Imatest plots demonstrate the characteristic curve obtained for a photograph of a Stouffer step wedge rendered with ACR into ProPhotoRGB. The default ACR TRC with a black point of 5 has a significant roll off in the shadows. Some say that the default of 5 is a bit aggressive and many photographers use a lower value. The TRC for ACR defaults with a black of 2 is shown along with ACR's linear TRC.

                                                                      I then converted the TIFF of the ACR rendering with default settings except for a black point of 2 into the color space of a typical printer, in this case the Epson 2200 premium luster paper. Relative colorimetric with black point compensation were used.

                                                                      It is interesting to note that the dynamic range for high quality results was reported by Imatest as 7.62 stops with the default ACR setting and 11.3 stops with the same settings except for a black point of 0.

                                                                      http://bjanes.smugmug.com/photos/472357094_sAg9X-O.gif
                                                                      • 32. Re: Exposure slider not entirely linear ?
                                                                        Jao vdL Adobe Community Professional & MVP
                                                                        >ACR uses a linear segment in the shadows even for aRGB

                                                                        Yeah that surprises me too. adobeRGB and ppRGB are not supposed to have the linear segment that sRGB has. When I switch between aRGB, ppRGB and sRGB in ACR and look at the histogram in the shadows, this does not seem to be the case and it appears that only sRGB has a different behavior in the shadows.
                                                                        • 33. Re: Exposure slider not entirely linear ?
                                                                          Tim Lookingbill Level 1
                                                                          I kind of suspected there was something different about these color spaces in ACR just by looking at the behavior of the live update of the histogram when adjusting the Exposure or Brightness slider.

                                                                          I noticed how it moves in a smooth swaying pattern as if trying to evenly distribute luminance while maintaining contrast proportions. Most moves in a true gamma space at least from my experience adjusting the middle gamma slider in Levels or apply a gamma shaped curve in Curves tends to clamp down the shadow region permanently no matter how bright you make the overall image until eventually the brightness reaches down to these areas but the highlites are practically blown or the overall image looks flat and washed out.

                                                                          I'm only assuming this to be the cause but for me this luminance adjusting behavior works great in squeezing out as much definition out of the shadows without kicking up posterization and noise. I've experienced the benefits of this trying to grab as much shadow detail as possible with very underexposed shots where the shadow detail reads 000RGB and turns to 10,10,10RGB with detail.

                                                                          To see what I mean apply a pinch curve to the 0-20RGB shadow regions in ACR's curve tool on a very underexposed image that has data that reads 000RGB and see how smooth you can come out of it. At least this is the case with my Pentax PEF's.
                                                                          • 34. Re: Exposure slider not entirely linear ?
                                                                            Panoholic Level 2
                                                                            > One might ask what is the practical significance of a linear segment in the shadows? I suspect that the significance is minimal since the shadows are clipped significantly in normal photograph use

                                                                            The shadow clipping occurs *before* the non-linear transformation (gamma application); otherwise you would not have any blacks in the resulting image, wasting a segment of the displayable dynamic range. Thus these are independent.

                                                                            Without the slope limitation, the first *two* levels of the range 0-4095 would "occupy" the first *eight* levels of the range 0-255.

                                                                            > adobeRGB and ppRGB are not supposed to have the linear segment that sRGB has

                                                                            The Adobe RGB 1998 specification *explicitely allows* for slope limitation, with the requirement that the slope limit do not exceed 1/32. Version 2005-05 mentiones, that Adobe's color conversion engine, PS and "other products from Adobe" impose a slope limit of 1/32.
                                                                            • 35. Re: Exposure slider not entirely linear ?
                                                                              Bill_Janes Level 2
                                                                              >The shadow clipping occurs *before* the non-linear transformation (gamma application); otherwise you would not have any blacks in the resulting image, wasting a segment of the displayable dynamic range. Thus these are independent.

                                                                              >The Adobe RGB 1998 specification *explicitely allows* for slope limitation, with the requirement that the slope limit do not exceed 1/32. Version 2005-05 mentiones, that Adobe's color conversion engine, PS and "other products from Adobe" impose a slope limit of 1/32.

                                                                              Gabor,

                                                                              The shadow clipping to which I am referring occurs when the tone curve with a blacks slider value greater than zero is applied. If the black slider is set to zero, I can detect no clipping.

                                                                              The actual May 2005 aRGB spec does not include a linear segment. See 4.3.1.2 and 4.3.4.2 of the spec. However, in Annex C of this document, a slope limit of 1/32 is allowed in a color converter. Thus ACR can legally use a slope limit when rendering an image into aRGB. This annex ends with the comment :

                                                                              "Note that the above slope limit is an implementation aspect, not an attribute of the Adobe RGB (1998) color space encoding. Different implementations may use different slope limits."

                                                                              http://www.adobe.com/digitalimag/pdfs/AdobeRGB1998.pdf
                                                                              • 36. Re: Exposure slider not entirely linear ?
                                                                                Panoholic Level 2
                                                                                Bill,

                                                                                > The shadow clipping to which I am referring occurs when the tone curve with a blacks slider value greater than zero is applied

                                                                                that's what I was thinking of. This adjustment occurs *before* the "gamma" application; consequently, you will have more zero pixel values after this adjustment than before, and you will have very low pixel values as well. Thus this step does not eliminate the effectivity of the translation in the very low range. Try it on an actual image: pick samples in different ranges with blacks = 0, and then increase blacks gradually. Blacks does not cause a "hole" at the dark end; they case a shift towards the left.

                                                                                > This annex ends with the comment

                                                                                We are talking about the very same document; however, its interpretation requires some reasoning, for it is nonsensical if taken literally. It says "the slope limit should not be greater than 1/32". Well, the lowest slope of the unmodified curve is 0.4547 (at x=1), thus the above restriction is plainly meaningless. The slope limit imposed by Adobe is in fact 32, not 1/32.

                                                                                Thus I interpret this paragraph as requiring, that the
                                                                                i greatest slope be not greater than 32.

                                                                                Note, that this is not very generous: sRGB limits the slope to 12.92, CIE L* requires a ramp with slope 9.033, Rec. 709 imposes a linear ramp with the slope 4.5. This means, that Adobe98 RGB "eats up" the low range by the shadows much quicker than other projections, even if the linear ramp is adopted.
                                                                                • 37. Re: Exposure slider not entirely linear ?
                                                                                  Bill_Janes Level 2
                                                                                  >Thus I interpret this paragraph as requiring, that the greatest slope be not greater than 32.

                                                                                  Gabor,

                                                                                  Yes, I think that is correct and the engineers at Adobe are not morons. The slope of 1/32 is for the inverse transfer function when one is going from the gamma encoded values back to linear for display or printing. The exponent in this direction is 2.2.

                                                                                  The transfer function going from linear to the gamma encoded aRGB, as with ACR rendering into a 2.2 gamma space, has a slope limitation of 32 and the exponent is 1/2.2
                                                                                  • 38. Re: Exposure slider not entirely linear ?
                                                                                    Jao vdL Adobe Community Professional & MVP
                                                                                    You learn something new every day!
                                                                                    • 39. Re: Exposure slider not entirely linear ?
                                                                                      Peter DL Level 1
                                                                                      > This means, that Adobe98 RGB "eats up" the low range by the shadows much quicker than other projections, even if the linear ramp is adopted.

                                                                                      Well said.


                                                                                      > Bill, Peter, I cannot speak for PS (since I don't work on it directly) but in Camera Raw (and Lightroom), the color spaces sRGB, Adobe RGB, and ProPhoto RGB all use a non-gamma segment in the shadows. For developers that are interested in the details (or anybody who is familiar with C++), please see the DNG SDK, source file dng_color_space.h.

                                                                                      Eric, - while Im not familiar with C++, Id be interested to understand if ACRs approach essentially corresponds to above discussed Slope limitation feature, or if it possibly goes beyond by imposing a more linear (flat angle) non-gamma segment in the shadows?

                                                                                      Many thanks for your comment.
                                                                                      Peter

                                                                                      --
                                                                                      1 2 Previous Next