Luminous Landscape Forum
Raw & Post Processing, Printing => Adobe Lightroom Q&A => Topic started by: pflower on July 09, 2009, 02:29:34 pm
-
Idle thought but ...
Any raw files generated by a digital camera are automatically imported by LR with a Brightness of +50. ( I am sure there is a way to disable that but that is not my question) Sometimes this is fine as far as I am concerned, sometimes the Brightness needs to be reduced. Depends upon the picture.
But what exactly does "Brightness" do? If I reduce Brightness to zero but then increase Exposure I can't get precisely the same effect.
So does anyone know what "Brightness" does as opposed to increasing exposure?
-
Exposure sets the white clipping point, brightness moves the midtones and tends to avoid clipping.
Personally, I tend to set exposure/blacks, put brightness/contrast to zero and use the tone curve to adjust the image.
-
Exposure sets the white clipping point, brightness moves the midtones and tends to avoid clipping.
Personally, I tend to set exposure/blacks, put brightness/contrast to zero and use the tone curve to adjust the image.
Thanks for that. Actually I do exactly the same thing, but I just wondered.
-
Exposure and Brightness are the same for the shadows, a linear scaling of the RGB channels. After that, Brightness starts to preserve the highlights by gradually compressing/expanding the high tones, while exposure linearly blows information in exactly the same way as the camera would do. Brightness never blows information, even if ACR's clipping warning activates; that happens just because the warning works with a threshold.
All this is much better seen by reverse-engineering the effect of the Brightness and Exposure sliders of ACR in terms of curves:
Exposure
(http://www.guillermoluijk.com/article/acrps/exposicion_acr.gif)
Brightness
(http://www.guillermoluijk.com/article/acrps/brillo_acr.gif)
By plotting them together it's easy to see that in the shadows and low mid tones:
* Exp +1EV equals Brightness +50
* Exp +2EV equals Brightness +100
* Exp +3EV equals Brightness +150
...
(http://www.guillermoluijk.com/article/acrps/exp_brillo_acr.gif)
Both controls are additive, so Exposure +4EV plus Brightness +150 will lift the shadows by 7 stops.
For more info on these and other adjustments, both in ACR and PS: HOW IMAGE LEVELS ARE AFFECTED BY ACR AND PS (http://www.guillermoluijk.com/article/acrps/index.htm). I was recently told that the brightness control in PS was lately (CS4, possibly CS3 too) redesigned, probably getting closer to ACR's brightness.
Regards
-
Once I asked about the exact relationships between the adjustments on the related Adobe forum. The answers were from useless blathering to plain nonsense; so I created a template to observe the effects. Now I redid it a bit better.
The raw file http://www.panopeeper.com/Download/SampleF...tments_Grey.dng (http://www.panopeeper.com/Download/SampleForACRAdjustments_Grey.dng) contains a graduated gray image (generated, not shot). If one loads this in ACR and leaves all adjustments as they are, then one gets a graduated grey image with a nice histogram, except in the black corner; I have not managed to smoothen it out.
Now, pick five-six sample points under each other, from the very top to the very botton and do different adjustments. The sample changes and the histogram show the effect.
ADDED: the color space must be sRGB
-
Exposure and Brightness are the same for the shadows, a linear scaling of the RGB channels. After that, Brightness starts to preserve the highlights by gradually compressing/expanding the high tones, while exposure linearly blows information in exactly the same way as the camera would do. Brightness never blows information, even if ACR's clipping warning activates; that happens just because the warning works with a threshold.
All this is much better seen by reverse-engineering the effect of the Brightness and Exposure sliders of ACR in terms of curves:
Guillermo,
An excellent demonstration--the best I've seen. Do you have a similar analysis for the recovery slider?
Bill
-
Nice work, Guillermo.
Another distinction: negative Exposure values will also invoke CR/LR's highlight recovery. Not true of Brightness; there is no connection between Brightness and CR/LR's highlight recovery logic.
-
I was recently told that the brightness control in PS was lately (CS4, possibly CS3 too) redesigned, probably getting closer to ACR's brightness.
The Brightness and Contrast adjustment was indeed changed in CS4. Both sliders were changed to behave more logically (not sure if its the same logic as ACR cause ACR is doing things in linear). CS4's adjustment has a "Legacy" checkbox in case you liked the old logic. Don't matter much to me cause I only ever used B&C on channels not images.
-
Another distinction: negative Exposure values will also invoke CR/LR's highlight recovery
What does this mean in clear-text regarding an image, which does not have clipped raw pixels? What about accurate ETTR and pulling back the "exposure" in the raw conversion?
-
Nice work, Guillermo.
Another distinction: negative Exposure values will also invoke CR/LR's highlight recovery. Not true of Brightness; there is no connection between Brightness and CR/LR's highlight recovery logic.
Eric,
Now that you have brought up the topic of highlight recovery, can you tell us when one should use the ACR recovery slider and negative exposure for images with burned out highlights? My own take is that one should use negative exposure if the whole image is too light from overexposure and the highlights are also burnt. In this case a linear correction is used. Recovery is best used for high dynamic range subjects where the overall exposure is correct, but the highlights are clipped because they exceed the dynamic range of the camera. In this latter case, the correction is non-linear, affecting mainly the highlights.
Bill
-
What does this mean in clear-text regarding an image, which does not have clipped raw pixels? What about accurate ETTR and pulling back the "exposure" in the raw conversion?
Gabor, if the image has no clipped raw pixels, then from CR/LR's point of view, there is no highlight recovery to perform. Therefore, using negative Exposure compensation in CR/LR does not perform highlight recovery.
-
Now that you have brought up the topic of highlight recovery, can you tell us when one should use the ACR recovery slider and negative exposure for images with burned out highlights? My own take is that one should use negative exposure if the whole image is too light from overexposure and the highlights are also burnt. In this case a linear correction is used. Recovery is best used for high dynamic range subjects where the overall exposure is correct, but the highlights are clipped because they exceed the dynamic range of the camera. In this latter case, the correction is non-linear, affecting mainly the highlights.
Bill, yes, that is how I would describe the distinction, too. As you know, in earlier versions of CR, there was only the Exposure slider, but no Recovery slider. Thus the only way to invoke highlight recovery was to use negative values for Exposure ... which, clearly, darkens the image overall. Hence, using Exposure purely for highlight recovery purposes has a very large side effect, which is that all pixels become darker, even ones in the shadows. This is a good thing if reducing overall exposure is one of your desired goals as well. However, if you like the midtones and shadows where they are already, then using negative Exposure will make them too dark. In the past, increasing Brightness was the way to bring them back.
Using the newer Recovery slider is rather like decreasing Exposure and increasing Brightness. When you increase Brightness, you cause the highlights to compress. This means that highlights will have less contrast. This is clearly indicated in Guillermo's Brightness plots.
-
Guillermo,
An excellent demonstration--the best I've seen. Do you have a similar analysis for the recovery slider?
Thanks Bill. I haven't that analysis since I am still using CS2 and a very old version of ACR (yes, I am so lazy to update the endlessly changing software).
But as Eric indicates in some of the previous posts, highlight recovery (activated over blown areas when exposure is set to negative values, or by the recovery sliders in modern versions of ACR), cannot be properly modelled in terms of curves since it is not such a simple operation, but 'something else'. Highlight recovery consists of some kind of replicating information from the non-blown channel/s in an area into the clipped ones, usually in neutral (R = G = B ) colours to preserve texture in a conservative non-coloured fashion. DCRAW's highlight recovery for instance tries to go further imitating the surrounding colour, but not always succeeds.
ACR left, DCRAW right
(http://www.guillermoluijk.com/tutorial/dcraw/recuperacion.jpg)
BTW Eric, can you give an explanation to the slightly unexpected behaviour in the deep shadows of the Exp/Brightness curves? thery are not 100% straight lines near 0. Has this some relation to the sRGB gamma curve?
Regards.
-
(yes, I am so lazy to update the endlessly changing software).
LOL, Guillermo. You certainly don't fit my interpretation of the word lazy. Your work and insight is very much appreciated, as is the rest of the brain trust that has already weighed in here in Eric, Jeff, Gabor, etc. This is quite a resource for the inquiring mind.
-
Using the newer Recovery slider is rather like decreasing Exposure and increasing Brightness. When you increase Brightness, you cause the highlights to compress. This means that highlights will have less contrast. This is clearly indicated in Guillermo's Brightness plots.
Another consequence of compression (beneath the non-linear section of such Brightness curve) is a loss of color saturation.
Unfortunately, the Recovery slider i.e. the combination of Recovery + Exposure do not seem to be free of this effect at all.
For the following test, the Red patch of a ColorChecker capture was driven close to clipping by means of the Exposure slider:
Exposure 2.4, Recovery 0, Brighness 0, Contrast 0, Curve Tab 0, Profile: base matrix.
HSB (pRGB)= 13/55/98 measured in Photoshop
Recovery was set to 25, then Exposure was raised again close to clipping:
Exposure 3.3, Recovery 25, Brighness 0, Contrast 0, Curve Tab 0, Profile: base matrix.
HSB (pRGB)= 12/40/98
Now Recovery was dropped, Exposure was reset and Brightness was raised (based on a mid gray for reference) to an overall similar output:
Exposure 2.4, Recovery 0, Brighness 40, Contrast 0, Curve Tab 0, Profile: base matrix.
HSB (pRGB)= 11/41/99
You may note that there’s a comparable major drop in highlight saturation.
Finally, I’m not using Recovery a lot, just a bit after Exposure, Brightness + Contrast and the Curve Tab are already set.
It’s still among my #1 reasons to edit in Photoshop (see true recovery (http://luminous-landscape.com/forum/index.php?showtopic=35412)).
Peter
--
-
Thanks Bill. I haven't that analysis since I am still using CS2 and a very old version of ACR (yes, I am so lazy to update the endlessly changing software).
But as Eric indicates in some of the previous posts, highlight recovery (activated over blown areas when exposure is set to negative values, or by the recovery sliders in modern versions of ACR), cannot be properly modelled in terms of curves since it is not such a simple operation, but 'something else'. Highlight recovery consists of some kind of replicating information from the non-blown channel/s in an area into the clipped ones, usually in neutral (R = G = B ) colours to preserve texture in a conservative non-coloured fashion. DCRAW's highlight recovery for instance tries to go further imitating the surrounding colour, but not always succeeds.
Guillermo,
Since you have not upgraded PS, I gather that ACR is not your main raw converter. IMHO, the improvements in ACR are alone are worth the upgrade price if you make much use of ACR.
As you indicate, highlight recovery with color data is complex, but I've done a few simple tests with a monochrome image using the green channel. These are not as exhaustive as your tests, but do give some useful data. I photographed a step wedge that was overexposed and rendered in ACR into sRGB using an exposure value of -0.5 EV to account for the BaselineExposure that ACR uses for my camera (personally, I do not like this "feature" of ACR). The top 3 steps (each 0.1 EV) are blown.
[attachment=15306:D3_Stouf_0003.png]
I then used exposure and recovery in an attempt to correct the overexposure. The results are shown graphically. As expected, exposure brings down all values by an equal value, whereas recovery affects the highlights and leaves step 10 alone.
[attachment=15307:Exposure.gif]
[attachment=15308:Recovery.gif]
Since a gamma curve has been applied by ACR, a log plot shows the relationships more accurately:
[attachment=15309:ExposureLog.gif]
[attachment=15310:RecoveryLog.gif]
Here is the raw histogram as shown by Rawnalize.
[attachment=15311:RawHistogram.gif]
Bill
-
Another consequence of compression (beneath the non-linear section of such Brightness curve) is a loss of color saturation.
Unfortunately, the Recovery slider i.e. the combination of Recovery + Exposure do not seem to be free of this effect at all.
Yes, that is a known characteristic of Camera Raw's tone curve implementation: when the tone curve is applied, areas that become darker have increased saturation; areas that become lighter have decreased saturation.
-
BTW Eric, can you give an explanation to the slightly unexpected behaviour in the deep shadows of the Exp/Brightness curves? thery are not 100% straight lines near 0. Has this some relation to the sRGB gamma curve?
Guillermo, I am not sure, to be honest.
What was your Point Curve setting when you made these plots?
-
Since you have not upgraded PS, I gather that ACR is not your main raw converter. IMHO, the improvements in ACR are alone are worth the upgrade price if you make much use of ACR.
You are probably right Bill, but the point is that I am not using any commercial RAW converter now. Time ago I found out that a totally neutral RAW development (like the one produced by DCRAW in 16-bit mode or ACR with _all_ settings to 0) is fine for me as a starting point. After that I just apply a couple of curves (some samples here (http://luminous-landscape.com/forum/index.php?s=&showtopic=20686&view=findpost&p=296852)), and believe me I don't miss anything else to fiddle with. I know this is a quite strange workflow, but it's OK for me. I am not a compulsive shooter so I can afford it.
I photographed a step wedge that was overexposed and rendered in ACR into sRGB using an exposure value of -0.5 EV to account for the BaselineExposure that ACR uses for my camera (personally, I do not like this "feature" of ACR).
I don't think there is any BaselineExposure in ACR. ACR simply can blow information when the exposure slider is set to 0 because of the white balance, which it seems to apply in a destructive way, i.e. with >=1 multipliers. To compensate for this overexposure due to WB, you need to add some negative exposure correction, not being this a problem at all. And it will not always be -0.5EV; some images may need -1EV, some others may be well served with -0.3EV depending on the WB applied and the particular exposure achieved in all three channels in the RAW data.
Eric my curve was set to totally linear for the reverse engineering tests. It seems as if ACR (and PS too) applies some extra lifting in the deep shadows when increasing the overall exposure, but no idea why.
Regards
-
BTW Eric, can you give an explanation to the slightly unexpected behaviour in the deep shadows of the Exp/Brightness curves? thery are not 100% straight lines near 0. Has this some relation to the sRGB gamma curve?
There's something "not quite right" with the ACR highlight recover too, which I think I'm seeing here and I've seen too (and reported to Eric). In the example below, using an ETTR capture, with both ACR and Raw Developer (awesome in terms of rendering quality), you can see that odd effect from using recovery. In the cement, it looks very similar the skin issue I see above.
(http://digitaldog.net/files/RD.jpg)
-
There's something "not quite right" with the ACR highlight recover too, which I think I'm seeing here and I've seen too (and reported to Eric
Have you seen this thread?
http://forums.adobe.com/message/1209993 (http://forums.adobe.com/message/1209993)
-
Have you seen this thread?
http://forums.adobe.com/message/1209993 (http://forums.adobe.com/message/1209993)
No. Thanks. These are also 5DMII shots.
-
I don't think there is any BaselineExposure in ACR. ACR simply can blow information when the exposure slider is set to 0 because of the white balance, which it seems to apply in a destructive way, i.e. with >=1 multipliers. To compensate for this overexposure due to WB, you need to add some negative exposure correction, not being this a problem at all. And it will not always be -0.5EV; some images may need -1EV, some others may be well served with -0.3EV depending on the WB applied and the particular exposure achieved in all three channels in the RAW data.
Regards
Guillermo,
ACR definitely uses a baseline exposure. One can determine the BaselineExposure by converting the raw file to DNG and reading the exif data. The pertinent extract from the EXIF is shown below. The BaselineExposure is discussed in the DNG specification (Link to PDF (http://www.adobe.com/products/dng/pdfs/dng_spec.pdf)).
I have found that if I use the BaselineExposure offset and set the ACR tone curve to linear, I get a fairly good representation of the raw file. For example here are the 10 steps in the image I showed previously. The table shows the ACR linear value for sRGB, the 14 bit raw value, the normalized raw value, and the normalized raw value converted to gamma 2.2 (=normalized raw ^[1/2.2] * 255). There is good agreement.
[attachment=15313:Table.gif]
Processing file D:\DataFiles\Nikon\D3\LowISO\Gabor\Jun092008_LowIso_0005.dng
Byte order: Least Significant Byte First
33432 Copyright ASCII Public Domain
50708 UniqueCameraModel ASCII Nikon D3
50721 ColorMatrix1 SRATIONAL 9336/10000 -3405/10000 14/10000 -7320/10000
14779/10000
2763/10000 -914/10000 1171/10000 8248/10000
50722 ColorMatrix2 SRATIONAL 8139/10000 -2171/10000 -663/10000 -8747/10000
16541/10000
2295/10000 -1925/10000 2008/10000 8093/10000
50727 AnalogBalance RATIONAL 1000000/1000000 1000000/1000000 1000000/1000000
50728 AsShotNeutral RATIONAL 503937/1000000 1000000/1000000 677249/1000000
50730 BaselineExposure SRATIONAL -50/100
50731 BaselineNoise RATIONAL 60/100
50732 BaselineSharpness RATIONAL 100/100
50734 LinearResponseLimit RATIONAL 100/100
50736 LensInfo RATIONAL 280/10 1050/10 35/10 45/10
50739 ShadowScale RATIONAL 1/1
50781 RawDataUniqueID BYTE
-
There's something "not quite right" with the ACR highlight recover too, which I think I'm seeing here and I've seen too (and reported to Eric). In the example below, using an ETTR capture, with both ACR and Raw Developer (awesome in terms of rendering quality), you can see that odd effect from using recovery. In the cement, it looks very similar the skin issue I see above.
I wouldn't call it issue, but just highlight strategy. When a single channel is clipped in the RAW file, no matter if the other two remain intact, ACR forces neutral (R=G=B ) highlights. It recovers texture, but always in B&W.
This image displays highlight recovery in an area with RGB channels clipped in the RAW data at different locations.
Consider the position of the red cross: only the G channel was blown there but the final image was gray in tha area after highlight recovery.
(http://img26.imageshack.us/img26/7711/raww.jpg)
ACR definitely uses a baseline exposure.
Thanks Bill, I didn't know about that. However even if that Baseline exposure were 0, you would get blown information because of WB, so an exposure correction would still be needed. To have the RAW developed without any exposure correction at all, is a fuzzy idea since WB means exposure correction. The only way to have your RAW files developed making sure no exposure correction is introduced, is to apply UniWB (1.0 multipliers) when developing, but this is a useless idea.
Regards
-
I wouldn't call it issue, but just highlight strategy. When a single channel is clipped in the RAW file, no matter if the other two remain intact, ACR forces neutral (R=G=B ) highlights. It recovers texture, but always in B&W.
Well whatever the strategy, I'll take that one built into Raw Developer. In the captures above, as well as other's I've produced, RD's strategy is always visually preferable.
-
Well whatever the strategy, I'll take that one built into Raw Developer. In the captures above, as well as other's I've produced, RD's strategy is always visually preferable.
I just feel curious Andrew, could you apply Raw Developer to my RAW file in the area I showed? and any in the thread is invited to try his own developer too.
Here is the link for download: RAW2 (http://www.guillermoluijk.com/download/raw2.cr2).
Regards
-
I just feel curious Andrew, could you apply Raw Developer to my RAW file in the area I showed? and any in the thread is invited to try his own developer too.
Here is the link for download: RAW2 (http://www.guillermoluijk.com/download/raw2.cr2).
Regards
At least on my end, that link provides a lot of garbage text. By all means, feel free to upload to my pubic iDisk:
My public iDisk:
thedigitaldog
Name (lower case) public
Password (lower case) public
Public folder Password is "public" (note the first letter is NOT capitalized).
To go there via a web browser, use this URL:
http://idisk.mac.com/thedigitaldog-Public (http://idisk.mac.com/thedigitaldog-Public)
-
During the weekends I am in the coast and have not enough DSL speed for heavy up/downloading, I'll upload it next Monday.
Anyway I didn't have any notice of that RAW file not working OK, anyone else tried?
-
I got it (using different browser).
What's interesting so far is, when using RD, turning on clipping overlay, I can fully remove it by altering its exposure to -2 stops, +28 for highlight recovery. In ACR, no adjustment removes clipping overlay! That said, I don't see the magenta effect on your image.
The question then becomes, do I attempt to make both converters roughly produce the same rendering appearance?
-
What's interesting so far is, when using RD, turning on clipping overlay, I can fully remove it by altering its exposure to -2 stops, +28 for highlight recovery. In ACR, no adjustment removes clipping overlay! That said, I don't see the magenta effect on your image.
The question then becomes, do I attempt to make both converters roughly produce the same rendering appearance?
If with 'clipping overlay' you mean the clipping warning in the RAW developer, I think it's just because of the different way both warnings works, not because one RAW developer performed better than the other. ACR is known to indicate blown areas when all three channels are RAW clipped no matter what controls are adjusted, giving up to recover any info on those areas because it's impossible. No idea how Raw developer works, but I am pretty sure it will not recover anything in the windows facing outside.
Magenta effect? what do you mean? as far as I know, everytime a magenta cast appears in the highlights is a software problem: the RAW developer is considering a saturation point for the RAW data higher than the real one. That produces a magenta cast where pure white should appear because of a lack of G values due to the white balance:
Wrong sat. point:
(http://www.guillermoluijk.com/tutorial/dcraw/sat1.jpg)
Correct sat. point:
(http://www.guillermoluijk.com/tutorial/dcraw/sat2.jpg)
There are considerations about the magenta issue in the DCRAW Tutorial (http://www.guillermoluijk.com/tutorial/dcraw/index_en.htm) (see 'SENSOR SATURATION LEVEL'), and in the article about the Fuji Super CCD (http://www.guillermoluijk.com/article/superccd/index.htm) (see 'TONO MAGENTA EN LAS ALTAS LUCES').
Could you please post the best image you manage to get from Raw Developer? I would like to see if it performs some colour interpolation.
Regards
-
Here's a quick and dirty comparison. What I tried to do was get the table to have the same degree of tonality in both but can't. In ACR/LR, I'm still not getting all the detail so to speak. I'm also seeing issues with smoothness albeit, without the magenta issues I saw on my 5DMII. It appears that ACR/LR boast saturation which I don't have a problem with. But look at the plants in the window and the beam of wood on the wall. I vastly prefer what I see in RD. Here's a JPEG but what I'll do is pop both in a folder in my iDisk. Again, this isn't super valid science because I'm sure I could spend more time with both products attempting to get a closer rendering match. There's a lot of other settings (sharpness, etc) that could be tweaked. But initially what I see is RD produces smoother, more "film like" rendering, and it appears to be able to pull out more highlight detail.
(http://digitaldog.net/files/RDvsACR.jpg)
-
If with 'clipping overlay' you mean the clipping warning in the RAW developer, I think it's just because of the different way both warnings works, not because one RAW developer performed better than the other.
Agreed. They just play differently. That said, I seem to be able to pull more detail out of RD.
Magenta effect? what do you mean?
Sorry. Its that color issue mentioned on the forum link which I should not have adopted. But in my cement shot (and other's I've produced), its some kind of color and smoothness issue. Often magenta.
-
Yes, it also seems to me that RD recovers more detail. I wouldn't dare to say, but I have a feeling that perhaps RD recovers more texture in areas where just one channel remained unclipped. Anyway it could just be a difference in contrast, i.e. the info is in the ACR image as well, but needs further processing. EDIT: on the left RD clearly obtained something while ACR seemed to produce pure white.
If you have some RAW file that clearly produces magenta to you, I'd like to analyse it.
Regards.
-
If you have some RAW file that clearly produces magenta to you, I'd like to analyse it.
Should be one on my iDisk in a folder called "For Drew" that produced the screen grabs above. There's two brackets, the one needing recovery is _MG_1247.CR2
-
EDIT: on the left RD clearly obtained something while ACR seemed to produce pure white.
Exactly and with very little effort. I can't get that data in ACR/LR no matter how I alter sliders or curves. And the area of the foliage outside the window with ACR looks tattered while the RD is smooth.
-
Exactly and with very little effort. I can't get that data in ACR/LR no matter how I alter sliders or curves. And the area of the foliage outside the window with ACR looks tattered while the RD is smooth.
I have extracted the RAW info, i.e. pure RAW data without demosaicing (the other image was an aproximation with neutral RAW development), taking only the R, G1 and B pixels of the Bayer matrix.
In that area where ACR displayed pure white, the R channel and only that channel, has some non blown info in the lines of the wooden floor. RD managed to take it into account and output it.
RAW R channel (RAW clipping blinking):
(http://img6.imageshack.us/img6/1584/28523742.gif)
RAW G1 channel (RAW clipping blinking):
(http://img40.imageshack.us/img40/6932/66024173.gif)
RAW B channel (RAW clipping blinking):
(http://img34.imageshack.us/img34/3152/58654287.gif)
-
Yes, that is a known characteristic of Camera Raw's tone curve implementation: when the tone curve is applied, areas that become darker have increased saturation; areas that become lighter have decreased saturation.
Understood regarding Brightness.
However, Recovery seems to do Luminosity recovery only,
even if the color information was/is in the Raw image (no real clipping) but just got affected e.g. by Brightness.
Peter
--
-
I don't think there is any BaselineExposure in ACR.
Regards
Guillermo,
I converted the raw file you referenced earlier to DNG and looked at the EXIF. It is for a Canon EOS 350D DIGITAL, but there is no BaselineExposure tag. Perhaps it doesn't exist for the 350D.
Bill
-
Understood regarding Brightness.
However, Recovery seems to do Luminosity recovery only,
even if the color information was/is in the Raw image (no real clipping) but just got affected e.g. by Brightness.
--
I looked at the behavior of recovery via negative exposure compensation in ACR using one of my overexposed step wedges and learned a bit about how it works. I selected an image that was overexposed by about 1 f/stop and split out the raw channels using the freeware astronomical program IRIS and the studied the resulting files in another freeware program ImageJ.
[attachment=15347:RawChannels.png]
In the green1 channel, the first four stops are blown out to white. Step 2 is intact in the red channel, as verified by the histogram. Step 1 is completely blown out in the blue channel, and step 2 is showing clipping in this channel.
I set the tone curve to linear in ACR and used a maximum negative exposure value of -4 EV, and the results are shown.
[attachment=15349:ACR_ScreenCap2.png]
In step 1, which is blown out in all channels, it is not possible to decrease the pixel value below 255. In step 2 there is valid date in the red channel and recovery is possible, but red=green=blue--i.e. the recovery is neutral as you and Guillermo mentioned. This behavior is also present in steps 3 and 4 which have valid data in the red and blue channels. Finally, in step 5 were all channels have intact data, color information appears. The white balance is slight off, and the RGB values are different.
When color information is lacking in one channel (in this the green channel), ACR has to guess, and using a neutral value is reasonable since extreme highlights often have little color. The Raw Developer program guesses that the color in areas with incomplete color information is similar to that in adjacent areas with intact channels.
This behavior can be useful in some instances, but it is better to expose so that channels are not blown. If the dynamic range exceeds that of the camera, two or more exposures can be made if the subject permits.
Bill
-
Guillermo,
I converted the raw file you referenced earlier to DNG and looked at the EXIF. It is for a Canon EOS 350D DIGITAL, but there is no BaselineExposure tag. Perhaps it doesn't exist for the 350D.
Bill
Mea Culpa--I forgot to convert to DNG. The baselilne offset is +0.25 EV
50708 UniqueCameraModel ASCII Canon EOS 350D
50709 LocalizedCameraModel ASCII Canon EOS 350D Digital
50730 BaselineExposure SRATIONAL 25/100
50731 BaselineNoise RATIONAL 133/100
50732 BaselineSharpness RATIONAL 150/100
50734 LinearResponseLimit RATIONAL 100/100
-
In step 1, which is blown out in all channels, it is not possible to decrease the pixel value below 255. In step 2 there is valid date in the red channel and recovery is possible, but red=green=blue--i.e. the recovery is neutral as you and Guillermo mentioned. This behavior is also present in steps 3 and 4 which have valid data in the red and blue channels. Finally, in step 5 were all channels have intact data, color information appears. The white balance is slight off, and the RGB values are different.
When color information is lacking in one channel (in this the green channel), ACR has to guess, and using a neutral value is reasonable since extreme highlights often have little color. The Raw Developer program guesses that the color in areas with incomplete color information is similar to that in adjacent areas with intact channels.
This behavior can be useful in some instances, but it is better to expose so that channels are not blown. If the dynamic range exceeds that of the camera, two or more exposures can be made if the subject permits.
Bill,
Agreed, however, my post was referring to a different aspect:
What happens to non-clipped highlight colors when we apply a.) Brightness, or b.) Exposure + Recovery, or c.) Brightness + Recovery via the respective sliders and positive settings thereof ?
Result: in all cases saturation is reduced and highlight colors may appear washed out.
In Photoshop for comparison this can be avoided by starting (likewise) with a 'linear conversion/rendition' from any suitable converter, and then by furnishing the tone curve with an inverted luminosity mask. A kind of prevention of 'damage' rather than belated recovery.
Peter
--
-
Should be one on my iDisk in a folder called "For Drew" that produced the screen grabs above. There's two brackets, the one needing recovery is _MG_1247.CR2
Out of interest I looked at _MG_1246.CR2 and _MG_1247.CR2 with Rawnalize. _MG_1246.CR2 is underexposed (green channels) by about 1 1/3 EV and _MG_1247.CR2 is overexposed by perhaps 0.5 EV, according to ETTR goals. The underexposure in the the first image when evaluated with ACR is partially masked by the baseline exposure of +0.4 EV that is used for the 5D Mark II. This is why I dislike the use of that offset. Unless you know about that offset and its value, it is difficult to evaluate exposure with ACR histograms.
[attachment=15371:Rodney1.gif]
The blue and red channel have unclipped data for recovery. Since the red multiplier for this camera is about 2, recovery should work for about to 1 stop overexposure for recovery.
[attachment=15372:Rodney2.gif]
-
Bill, the baseline exposure is a value that is added to the RAW developer exposure adjustment value, or the RAW developer adjustment is set to the baseline exposure as a starting default value than can then be changed?
I.e., if you set 0.0EV in ACR you are applying the baseline exposure, or you are just cancelling it?
Regards
-
Bill, the baseline exposure is a value that is added to the RAW developer exposure adjustment value, or the RAW developer adjustment is set to the baseline exposure as a starting default value than can then be changed?
I.e., if you set 0.0EV in ACR you are applying the baseline exposure, or you are just cancelling it?
Regards
This is what the DNG specification says: "Because of these differences, a raw converter needs to vary the zero point of its exposure compensation control from model to model. BaselineExposure specifies by how much (in EV units) to move the zero point. Positive values result in brighter default results, while negative values result in darker default results.". However, this seems backwards to me. The BaselineExposure for the D3 is -0.5. However, ACR renders the image too light with no exposure compensation, and I have to use a negative exposure adjustment of -0.5 EV to get correct results. By correct, I mean a sensor saturation of 18% when exposure is made according to the light metered exposure, resulting in a pixel value of 118 in a gamma 2.2 space. With your 350D, the BaselineExposure is +0.25, so I would recommend trying an exposure compensation of +0.25 EV for correct tone placement.
The BaselineExposure for the 5D MII is +0.4. According to an article (http://www.libraw.org/articles/Canon-5Dmk2-headroom.html) on Iliah Berg's web site, this camera places mid gray at 6.5% saturation. He has indicated that this is because of certain limitations of the sensor (see here (http://forums.dpreview.com/forums/read.asp?forum=1021&message=32246111). For the 5D MII one could try using +0.4 EV exposure compensation in ACR. However, this might not be enough if Iliah's obsrvations are correct.
Hopefully, Eric Chan is still following this thread and can comment.
Bill
-
This is what the DNG specification says: "Because of these differences, a raw converter needs to vary the zero point of its exposure compensation control from model to model. BaselineExposure specifies by how much (in EV units) to move the zero point. Positive values result in brighter default results, while negative values result in darker default results.". However, this seems backwards to me. The BaselineExposure for the D3 is -0.5. However, ACR renders the image too light with no exposure compensation, and I have to use a negative exposure adjustment of -0.5 EV to get correct results. By correct, I mean a sensor saturation of 18% when exposure is made according to the light metered exposure, resulting in a pixel value of 118 in a gamma 2.2 space. With your 350D, the BaselineExposure is +0.25, so I would recommend trying an exposure compensation of +0.25 EV for correct tone placement.
The BaselineExposure for the 5D MII is +0.4. According to an article (http://www.libraw.org/articles/Canon-5Dmk2-headroom.html) on Iliah Berg's web site, this camera places mid gray at 6.5% saturation. He has indicated that this is because of certain limitations of the sensor (see here (http://forums.dpreview.com/forums/read.asp?forum=1021&message=32246111). For the 5D MII one could try using +0.4 EV exposure compensation in ACR. However, this might not be enough if Iliah's obsrvations are correct.
Hopefully, Eric Chan is still following this thread and can comment.
Bill
Perhaps I'm misunderstanding this, but if the baseline exposure for a camera is +0.25, doesn't that mean ACR is automatically adding 0.25 exposure compensation "behind the scenes" when set to 0 and that you should set it to -0.25 to cancel it out and see the actual RAW exposure?
If you were to add +0.25, wouldn't that be 0.5 over what the RAW exposure actually is?
-
Perhaps I'm misunderstanding this, but if the baseline exposure for a camera is +0.25, doesn't that mean ACR is automatically adding 0.25 exposure compensation "behind the scenes" when set to 0 and that you should set it to -0.25 to cancel it out and see the actual RAW exposure?
If you were to add +0.25, wouldn't that be 0.5 over what the RAW exposure actually is?
You do understand it right; the baseline exposure correction is applied automatically. What worse is the fact, that it is not only automatic, but clandestine. The slider does not show it, and it is not described anywhere (the amount depends on the camera model and often on the ISO as well).
For example Canon's Highlight Tone Protection option reduces the effective ISO from the displayed one by a full stop, i.e. the shot is underexposed, with the aim of increasing the intensity in the raw conversion, except for the very highlights. ACR increases the intensity by applying +1 EV (but that's it, ACR does not mimic the in-camera and DPP adjustment of the curve). The user does not see this adjustment at all; thus (s)he believes the exposure was correct, which is misleading pure.
MFDB users are regularly misled by ACR into believing, that they understand how their camera works and are exposing correctly.
-
Perhaps I'm misunderstanding this, but if the baseline exposure for a camera is +0.25, doesn't that mean ACR is automatically adding 0.25 exposure compensation "behind the scenes" when set to 0 and that you should set it to -0.25 to cancel it out and see the actual RAW exposure?
If you were to add +0.25, wouldn't that be 0.5 over what the RAW exposure actually is?
To answer your question with an example for which I have data, let's use the D3, which has a BaselineExposure value of -0.5 EV as shown in a previous post. If that exposure adjustment were automatically used behind our back so to speak, it would darken the image. I photographed a gray card according to the camera exposure meter reading. The resulting pixel value in the raw file is approximately 1820 or 94 in terms of 8 bit sRGB as shown in this Rawnalize capture. (The Nikon metering is the the equivalent of 12.5% reflection, not 18%. If you want 18% saturation for mid gray you need to add 0.5 EV).
[attachment=15429:Rawnalize.png]
Setting the ACR tone curve to linear and leaving the exposure slider at zero, the sRGB pixel value is 110, which is too light. The BaselineExposure of -0.5 lightens the image rather than darkening it.
[attachment=15430:ACR_LinearExp0.png]
Setting exposure to -0.5 EV gives an sRGB value of 96, very close to the predicted result of 94:
[attachment=15431:ACR_Line...inusHalf.png]
Using ACR defaults gives an image which is much too light. Mid gray is 159. If you didn't know better, you would think that the camera was overexposing. Such is a common complaint on the Nikon forums. Using an exposure value of -0.5 EV in ACR, I get an sRGB value of 126 which is more reasonable (mid gray is 118 in sRGB).
[attachment=15432:ACR_Default.png]
-
What worse is the fact, that it is not only automatic, but clandestine.
hahaha clandestine, I really love how you use the language sometimes. Still, Gabor and Bill, I don't think there is a clear way to speak about the _actual_ exposure achieved in the camera, since the white balance means a change in exposure in at least 2 of the three channels.
If we want to be strict about the _actual_ achieved exposure and the 18% gray story (which is basically useless, but interesting from a technical point of view), white balance should be entered into the equation.
However I always considered the discussion about the 18% and the actual exposure quite useless for several reasons:
- Camera exposure settings are not fine tuned, they are rounded to 1/3EV (this means an average error of at least 1/12EV=0.08EV, and max error of at least 1/6EV=0.17EV).
- Any RAW development and/or processing will involve tonal shifts (any level adjustment means a tonal shift), that can both corrrect or ruin the 18% criteria. In fact most RAW developers like ACR calculate the initial settings to obtain a pleasant image, not using any 18% criteria.
- The precise point from saturation at which each digital camera locates a light metered area is irrelevant. Cameras with a large gap from metering to sat can be used with +EC to compensate for this. Cameras with a reduced highlight headroom from metering to sat can be used with a systematic -EC compensation.
At least for RAW shooters, the only important thing to know is how your camera behaves: i.e. how many EV it allows from metering to sat, and act accordingly depending on your needs and situation.
These are the RAW log2 histograms from shots made over a tungsten uniformly lighted surface, according to camera metering for 350D and 5D:
350D
(http://www.guillermoluijk.com/article/spot/histomax350d.gif)
5D
(http://www.guillermoluijk.com/article/spot/histomax5d.gif)
Both cameras in their 3 RAW channels produce a RAW level well below (more than 1EV) the expected [18% -> -2.5EV] or the [128 in a gamma 2.2 encoding -> -2.2EV], so even after a positive white balance it won't reach the expected middle gray (128) in a 2.2 gamma encoding. But it's unimportant.
Regards
-
Still, Gabor and Bill, I don't think there is a clear way to speak about the _actual_ exposure achieved in the camera, since the white balance means a change in exposure in at least 2 of the three channels.
If we want to be strict about the _actual_ achieved exposure and the 18% gray story (which is basically useless, but interesting from a technical point of view), white balance should be entered into the equation.
I must disagree with our esteemed colleague. For exposure, one usually takes the value of the green channel for which the white balance multiplier is unity. With daylight exposure, the red and blue channels are underexposed, but the WB multipliers bring them up to the level of the green channels.
However I always considered the discussion about the 18% and the actual exposure quite useless for several reasons:
- Camera exposure settings are not fine tuned, they are rounded to 1/3EV (this means an average error of at least 1/12EV=0.08EV, and max error of at least 1/6EV=0.17EV).
- Any RAW development and/or processing will involve tonal shifts (any level adjustment means a tonal shift), that can both corrrect or ruin the 18% criteria. In fact most RAW developers like ACR calculate the initial settings to obtain a pleasant image, not using any 18% criteria.
- The precise point from saturation at which each digital camera locates a light metered area is irrelevant. Cameras with a large gap from metering to sat can be used with +EC to compensate for this. Cameras with a reduced highlight headroom from metering to sat can be used with a systematic -EC compensation.
At least for RAW shooters, the only important thing to know is how your camera behaves: i.e. how many EV it allows from metering to sat, and act accordingly depending on your needs and situation.
These are the RAW log2 histograms from shots made over a tungsten uniformly lighted surface, according to camera metering for 350D and 5D:
Both cameras in their 3 RAW channels produce a RAW level well below (more than 1EV) the expected [18% -> -2.5EV] or the [128 in a gamma 2.2 encoding -> -2.2EV], so even after a positive white balance it won't reach the expected middle gray (128) in a 2.2 gamma encoding. But it's unimportant.
If you use an automated exposure mode such as aperture priority, the camera shutter speed is adjusted to the exact value needed and not in the 1/3 stop increments indicated on the camera LCD. To test this, I took exposures with my Nikon D3 of a white wall which was illuminated unevenly, using a 300 mm lens to select small areas which were evenly illuminated over a small area. I used ACR with a linear tone curve and rendered into sRGB using the required exposure offset of -0.5 EV and read a small central area of the images using a mask in Photoshop so I would be reading the same area in each shot.
The indicated exposures were 1/15, 1/13, and 1/10 sec. The sRGB readings were 107.41, 107.95 and 107.29, indicating fine control of exposure. Rawnalize gave a gamma 2.2 value of 109 and a 14 bit raw value for green of 2418. The camera was giving a saturation of 15%, about 2.75 stops below 100% saturation. The ISO saturation standard leaves 0.5 EV of headroom, so a saturation of 12.5% would be expected. ACR with the required exposure offset and a linear tone curve gave an accurate result. However, I do agree that you do need to know your camera and how it exposes. In addition, you need to know your raw converter and the tone curve used by the camera JPEG engine. The Nikon D3 leaves slightly more than 0.5 EV of headroom and then uses a hot tone curve to bring up the value. With Capture NX and the Standard Picture Control, I got an sRGB value of 161. NX closely approximates the JPEG engine of the camera. With ACR defaults and the Adobe Standard camera profile, the sRGB value was 153.
BTW, mid gray in gamma 2.2 is 118, not 128. See Bruce Lindbloom's companding calculator for L* = 50. The normalized pixel value is 0.4635, or 118.2 in 8 bit notation.
Best regards,
Bill
-
I don't think there is a clear way to speak about the _actual_ exposure achieved in the camera, since the white balance means a change in exposure in at least 2 of the three channels
At the moment we are inspecting the raw data, the term "exposure" as the combination of aperture and shutter time plays no role any more. From here "exposure" is the pixel value. The only important aspect is if the pixel value is reliable, i.e. not clipped and not in the very noisy range; everything else is raw processing. I personally don't care the very least for the "correct exposure", I am looking at the raw-like histograms and adjust the exposure as I see fit (assumed that the setting allows for the required adjustment).
The clandestine adjustment of the "exposure" ruins that world by falsifying the result. Extreme example: the Phase One P45+. Its ISO steps are supported by analog gain (it plays no role in this aspect, how useful the increased analog gain is). Consequently, increasing the ISO and reducing the exposure yields quasy the same pixel values. Still, ACR adds one full stop to ISO 400 and two stops to ISO 800. The result is, that you believe you have horrendeously overexposed the shot. You reduce the exposure and get a "correct" shot - in fact you are heavily underexposing. The result is, that this rubbish (the clandestine adjustment) costs the photographer a lot.
Or take Canon's Highlight Tone Priority - the vast majority of phographers do not know, that the only effect of HTP when shooting raw and using ACR is, that they get more noise; this too is the result of BaselineExposure.
If you use an automated exposure mode such as aperture priority, the camera shutter speed is adjusted to the exact value needed and not in the 1/3 stop increments indicated on the camera LCD
I have a hard time to believe, that the camera sets the shutter speed to the exact value required to the correct exposure. Though I saw some indication, that at least some of the Nikon cameras can set exposure in 1/6 EV implicitely.
mid gray in gamma 2.2 is 118, not 128. See Bruce Lindbloom's companding calculator for L* = 50. The normalized pixel value is 0.4635, or 118.2 in 8 bit notation
Why would one define mid-grey as L*=50, instead of B=50%? Some people are not confronted with Lab in their entire photographic life.
-
The clandestine adjustment of the "exposure" ruins that world by falsifying the result.
I agree entirely and think that the baseline exposure that Adobe uses is counterproductive for ETTR exposure, because it hides the actual exposure.
I have a hard time to believe, that the camera sets the shutter speed to the exact value required to the correct exposure. Though I saw some indication, that at least some of the Nikon cameras can set exposure in 1/6 EV implicitely.
My assumption was that the shutter speeds in aperture priority mode remain in the analogue domain and do not increment in 0.33 EV units. My testing was not extensive, but my results are consistent with this assumption. What if you have the exposure increment set to 0.5 EV via a custom function? One could perform more testing and determine the granularity of the shutter speed, using the raw value as a measure of the exposure as you indicated previously.
Why would one define mid-grey as L*=50, instead of B=50%? Some people are not confronted with Lab in their entire photographic life.
Because L* = 50 is mid gray on the perceptual scale. A few tests in Photoshop indicate that B = 50 is L* = 54 in sRGB and 61 in ProPhotoRGB.
Wikipedia: "Because HSL and HSV are simple transformations of device-dependent RGB models, the physical colors defined by (h, s, l) or (h, s, v) triplets depend on the colors of the red, green, and blue primaries of the device or of the particular RGB space, and on the gamma compression used to represent the amounts of those primaries."