Girod,
For the benefit of other website visitors that haven't had the opportunity to perform these tests themselves, I thought I would post the results of my "white towel" exposure testing. Please let me know if your results agree with these.
Again, the purpose of this phase of my testing was to determine where "Zone VII" is on the exposure scale. Zone VII is defined as "the highest zone that will still retain good details" (such as textured snow, finely patterned white material, etc.) When using a modified Zone System to base our exposures on in the field, we need to understand how to optimize our exposures to "retain good detail" in the important highlight areas of the image.
We used a "white towel" as our target because it has plenty of fine detail and texture that will allow us to judge when we have successfully exposed the image at the "highest zone to retain good detail." Since we had already determined by examining the raw data that the saturation point for the D3/D700 sensor is +3.0 EV from the "mid-tone" meter reading , I used that as the "upper end" of the test range for this series of images.
I set up the camera on a tripod, focused on the white towel in even diffuse lighting on the floor below the tripod, set an aperture of f8, closed the eyepiece shutter, spot metered and set the exposure at +3.0 EV (I checked the meter by "defocusing" the image to ensure it wasn't effected by uneven lighting on the towel), and took the exposure using "mirror up" and a cable release. I repeated this process in -1/3 stop increments down to +1.7 EV above mid-tone.
I then opened each image into Capture NX and used the "highlight protection" slider to pull the histogram back down into the image in an attempt to recover the highlights. I used the "show lost highlights" view to judge how much I needed to move the "highlight protection" slider. I set the highlight protection slide at the point where the "show lost highlights" view no longer displayed any clipping. I then examined each image for the level of detail present.
Here is the +3.0 EV image, with the highlight protection slider set at 80/100. Both the histogram and "show lost highlights" view imply that all the highlights have been recovered; however, examination of the image at 100% view tells a different story:
(http://spiritofphotography.com/coppermine/albums/userpics/10001/white_towel_%2B3.jpg)
+3.0 EV, "highlight protection" applied
(http://spiritofphotography.com/coppermine/albums/userpics/10001/white_towel_%2B3_zoomed.jpg)
+3.0 EV, "highlight protection" applied, zoomed to 100% (cropped)
Here is the +2.7 EV image, again with the highlight protection slider set so that the histogram and "show lost highlights" view imply that all the highlights have been recovered. I applied a white balance adjustment to this image, so the highlight protection slider had to be set to 85/100 to remove all the "clipping." Here's the "detail" available in this image:
(http://spiritofphotography.com/coppermine/albums/userpics/10001/white_towel_%2B2_7_zoomed.jpg)
+2.7 EV, "highlight protection" applied, zoomed to 100% (cropped)
Here is the +2.3 EV image. Same process, highlight protection slider set at 31/100:
(http://spiritofphotography.com/coppermine/albums/userpics/10001/white_towel_%2B2_3_zoomed.jpg)
+2.3 EV, "highlight protection" applied, zoomed to 100% (cropped)
Here is the +2.0 EV image. Same process, highlight protection slider set at 6/100:
(http://spiritofphotography.com/coppermine/albums/userpics/10001/white_towel_%2B2_0_zoomed.jpg)
+2.0 EV, "highlight protection" applied, zoomed to 100% (cropped)
Adobe Camera Raw has traditionally done a better job of recovering highlights than CaptureNX, so out of curiosity I opened the +2.3 EV file in Adobe Camera Raw 5.3, set the exposure compensation to -1.0, and performed a click balance white balance adjustment. Again the histogram and "clipped highlights" display indicated that all highlights had been "recovered." Here's the detail in the file:
(http://spiritofphotography.com/coppermine/albums/userpics/10001/white_towel_%2B2_3_zoomed_ACR_5_3_whitebal_and_-1_0expcomp.jpg)
+2.3 EV, processed in ACR 5.3, -1.0 EV on exposure slider, zoomed to 100% (cropped)
The ACR conversion does appear to recover highlights better than the NX2 conversion of the +2.3 EV file, providing a smoother tonal transition in the areas that are recovered; however, this rendition falls way short of the detail contained in the +2.0 EV image from NX2. I tried converting the +2.7 EV file in ACR 5.3, and the rendition was unacceptable, showing the same non-recoverable artifacts as the NX2 renditions. It appears that ACR 5.3 will provide 1/3 stop additional "margin" in recovering highlights, although at the cost of significant image detail.
Someone that has been following this discussion might now be asking "if my sensor doesn't saturate or "clip" until +3.0 EV, why in the heck do I need to expose highlights at +2.0 EV in order to retain detail in the highlights?" The answer is that those "details" span a 1.3 stop range in the image (in pretty much a "bell curve" type of distribution for the image of the towel). Looking at the +2.0 EV image of the towel again, it certainly doesn't appear to our eye that this is a 1.3 stop range of brightness, but it is. This is consistent with just about any "real world" scene where you want to "retain detail" in the highlights.
Lets look at the raw data histogram for the EV +2.3 image (without whitebalance multipliers applied) in Rawnalyze:
(http://spiritofphotography.com/coppermine/albums/userpics/10001/%2B2_0_EV_Rawnalyze.jpg)
White towel spot metered at +2.3 EV, raw data histogram from Rawnalyze
You can see from this screen shot of the raw data histograms that even when metering to expose the image at 0.7 EV below the saturation point of the sensor, there is still a substantial percentage of the green and blue channels that are "clipped" in the raw data. (The "saturation point" of the sensor is annotated by the dotted yellow line on the right side of the histogram.) Lowering the exposure to +2.0 EV moves this "clipped" data into the range that can be accurately recorded by the sensor.
Therefore, if my goal is to place "important" highlights into Zone VII, "the highest zone that will still retain good details," then I will expose these highlights at +2.0 EV using the camera's spot meter. This is consistent with the Modified Zone System that I have been using for three generations of digital cameras, including Nikon, Canon and Kodak digital SLRs.
Keith
Girod,
A couple of thoughts:
1) There is a bit of inconsistency between our results. One reason is because we were testing under light with different color temperature (shade vs direct sunlight). However, this doesn't explain all of the differences we are seeing. I believe there might also be a bit of inconsistency both due to our interpretation of the results and perhaps a misunderstanding of how to "apply" white balance in Rawnalyze.
Here is a screen shot of the histograms displayed in Rawnalyze for the +2.0 EV image with white balance applied:
(http://spiritofphotography.com/coppermine/albums/userpics/10001/%2B2_0_EV_Rawnalyze_WB_applied.jpg)
"white towel" exposure test at +2.0 EV, histogram in Rawnalyze with "selection" WB applied
Using the "composite" view, I made a large selection (right click and drag) from the center of the image, then clicked Ctrl + W to apply a white balance correction to the image using the "selection" to calculate a neutral white balance rendition. The "hash marks" on the horizontal axis of the Rawnalyze histogram display indicate 1/3 stop increments. With white balance applied the red and green channels are essentially just below clipping and the blue channel is clipping slightly.
Here is a screen shot of the Rawnalyze histogram of the +2.3 EV image with white balance applied:
(http://spiritofphotography.com/coppermine/albums/userpics/10001/%2B2_3_EV_Rawnalyze_WB_applied.jpg)
"white towel" exposure test at +2.3 EV, histogram in Rawnalyze with "selection" WB applied
Note that at +2.3 EV, with the "proper" white balance applied, all three color channels indicate clipping. In my case (because I was shooting in "cool" light) I was unable to recover the detail in the towel because both the green and blue channels were clipping in the raw data.
I hope you don't think I'm harping on this too much. I really want to make sure we are on the same page here because the most "unforgiving" aspect of digital photography is when data is "clipped," i.e., the sensor is saturated. If the data is clipped, the software can "interpolate" (guess at) the data to try to reconstruct the missing detail, but this interpolation is a poor approximation of the "real" data. If it isn't important to display this portion of the image with all of the detail that is available in the original scene, then that is OK, but it doesn't belong in Zone VII if it isn't important.
It is widely accepted by followers of the Zone System that "perceivable detail" belongs in Zones III through VII. For a high-contrast scene (using a digital camera), you would expose to place important highlight detail in Zone VII and then adjust mid-tone and shadow placement as needed during post-processing. We need to understand how to expose correctly to make that happen on a consistent basis. Mid-tone and shadow placement are less important during the initial image capture because we have more latitude in adjusting placement of those tones in post-processing.
2) I think the results of our tests under different color temperatures of light reinforce that our success at interpolating data from clipped highlights will vary depending on the "warmth" or "coolness" of the light source. With "direct sunlight," as long as only one channel is clipped, the software can interpolate the data in the other two color channels to "reconstruct" a portion of the missing data in the clipped channel. The success of this interpolation will depend on how consistent the data was in "real life" in the clipped channel. With cooler light, we are in danger of clipping both the green and blue channels at the same time. The challenge is to know when we are truly clipping the blue channel, and not be fooled by the white balance multiplier being applied to the data during the in-camera or CaptureNX2 processing.
A general rule of thumb for the D3 and D700 is that in "shade" under a blue sky, the blue channel will clip at about the same time as the green channel. The blue channel coefficient (for the D3/D700) for "shade" is 1.0859, so if you see blue channel clipping when you have "shade" set for white balance (or when you are using auto WB and shooting in shade under a blue sky) then you should expect that the "raw" blue channel data is truly clipping.
There are differing opinions with respect to how "hot" we can expose the highlights and still obtain usable results. It is true that in some situations, you can expose (Zone VII) highlights at +2.3 EV, or even +2.7 EV, and still obtain usable results by using "highlight protection" or other methods to recover the highlights. My goal is to be able to consistently obtain optimum exposures in challenging lighting conditions. I know that I have a lot of latitude in the shadows with the D3 (and D700) and so adopting a "conservative" method of exposing the highlights has served me very well. Exposing highlights with visible detail at +2.0 EV will almost always enable me to produce a viable image from the recorded data. Exposing any hotter than this is "treading on thin ice."
I would very much like to continue this discussion, but I need to take a break now to finish getting ready to teach a workshop in Crested Butte, Colorado. Please check back with me on Tuesday of next week and we can continue our discussion.
Keith