I'm working on a lot of tests for an article on expose to the right. To make sure I'm clear here, its important to differentiae highlight recovery from simply altering the default rendering controls based on expose to the right (ETR). In Lightroom and CR, I'm able to dial down the exposure slider and fully retain very bright specular highlights. The best target I've got has a luminance value I measured with a Spectrophotometer of 99.8 L*Star. That's really white!
Would that be the Babelcolor White target? I measured mine with the EyeOne Pro, and got an almost constant 98.7 L* up to 100.06 L* (but 99.9 L* for most wavelengths) over the full 390-710 nm spectrum. It's an amazing tool for both exposure to the right, and white-balancing right up to the 1/3rd stop below clipping of the Raw data.
I can over expose over what the external meter suggests 1.5 stops and using (in the case of ISO 100) -1.47 exposure setting in Lightroom, keep this target just below 100% value. There is full tonal data in all three channels even 'over exposing' this amount.
This is the point where most evaluations become hard to follow, due to what "the meter suggests". When using an external incident light meter, one could assume that it is calibrated to something like 12.5% or 18% mean luminance, depending on manufacturer. A similar assumption can be used for spotmetering (either internal to the camera or external), or unweighted averaging metering. Any 'evaluative' or other semi-intelligent metering will be a guess as to how it reacts, perhaps even on a uniformly lit surface.
My 'simple' method to determine non-clipped exposure latitude (number of exposre levels / EV / stops above metered average) is to expose a uniformly lit flat surface as suggested by the exposure meter (one can verify between metering modes to make sure), and then bracket up in 1/3rd stop increments (some cameras only allow half stops). Usually somewhere between 3 and 4 stops exposure above average, all 3 color channels will clip into oblivion. Ideally that should be checked on the Raw file data,
before CFA demosaicing and color-balancing. A free tool like IRIS will allow to do that, it produces a grayscale file from Raw and allows to separate the R, G1/G2, and B sensel data.
That uniformly lit surface by the way, can also be created by placing a piece (or stack) of opaline glass flush to the front of the lens (approx. f/5.6 and 50 - 100mm focused at infinity), and point it at something reasonably uniformly lit. Then the cropped center (upto a couple of hundred pixels square) of the image will be relatively unaffected by lens vignetting / light fall-off, and lighting is very diffuse.
There IS the ability of some raw converters to take one of three color channels where the other two are blown out completely and rebuild highlight data. That's not what I'm seeing here. The 1.5 plus exposure has full tonal data in all three channels.
Understood, you are not referring to highlight recovery. However, when we assume L* 42 for 12.5% reflection (or L*50 for 18.4%) and L* 99.7 for 99.2% reflection, that would give 2.99 (or 2.43) stops latitude above average exposure, before clipping, with 1/3rd stop exposure accuracy. Anything beyond that clipping point will be the result from Raw conversion or (non-linear) tonemapping if not looking at the Raw data.
Now, depending on the color of the light, and the native color balance of the sensor array, there often
is a bit of non-clipped info in one or two of the channels, and that can be exploited for highlight detail or for a small boost in overall exposure (depending on the subject contrast). That also means that for exposure-to-the-right of medium to low contrast subjects we can increase average exposure which will benefit the overall signal to noise ratio.
It looks over exposed when viewed using a default LR or CR rendering as you'd expect. In the case of the ISO 100 image, simply moving the exposure slider down as mentioned above, produces a rendering that now appears nearly identical to the image shot at 'normal' exposure. Only when you view the two images rendered for a normal appearance do you see the quality benefits of the initial 'over exposed' image in the shadows (less noise, much smoother). This is due to the linear nature of raw data. Half of all levels are contained in the first stop of highlight data.
Yes, especially lower contrast images will look overexposed, but they do benefit the most from pulling exposure at the Raw conversion stage.
I should also bring these brackets into another converter (probably my gold standard, Raw Developer). But I suspect with a similar exposure compensation there, I will be able to produce a rendering that appears 'normal' and superior from the image the meter 'thinks' is 1.5 stops over exposed.
Please make sure that the 'over' exposure you are going to mention in your final article is related to a well understood 'normal' exposure. My suggestion is to specify the exposure latitude above average exposure, because that can be determined accurately and independently by everybody for their own camera and specific metering calibration.
I did the bracket up to 2 stops over. In the case of the 5D, 2 stops really IS over exposed. The sensor resulted in overload (full sensor saturation). No amount of alteration of the rendering controls produce less than 100% in the specular highlights. The data is gone.
Depending on you exposure metering(!), and assuming a calibration for medium gray at 12.5% luminance, your White target should almost exactly fit in the latitude, but maybe I misunderstand what you are saying. That doesn't mean that we disagree about the huge potential that e.g. Lightroom/ACR adds to image quality, it does. It also stresses the point to expose for the whites, to avoid clipping beyond detail repair and maximize signal to noise in single exposure shots. That is also the point that the Camera's clipping indicator on the LCD may be of use, but that depends on what it is based on (luminance, unweighted RGB, or a single channel).
Looking forward to the final article,
Bart
