Recommended Reading: "Optimizing Exposure," Luminous Landscape.com

[keithsnell]

From time to time I'll post links to "recommended reading," or references that I think do a fairly good job of explaining photographic concepts.  The attached link leads to an article that discusses why the photographer may want to consciously modify the camera's recommended exposure in order to "optimize" the exposure.

I would add one thing to what the author says.  In the article he states:

"With a typical DSLR, when photographing a red flowers under natural daylight, the LCD histogram will typically show the red channel as blown out. This doesn't tell whether the native raw red channel is actually blown. So one doesn't know whether to increase the exposure for ETTR [expose to the right], or reduce it. The natural reaction of most users is to say, "Uh oh, I'm gonna blow the red channel in these flowers, so I better reduce the exposure till the red histogram doesn't look blown out anymore." Unfortunately, that's almost always the wrong thing to do. In fact, the red channel (in the raw data) rarely clips on a typical DSLR with a normal daylight exposure, because the red sensitivity is very low (about 1.5 stops darker than green). If one was to reduce exposure till the red histogram no longer showed clipping, then the actual raw red channel would be very underexposed with a poor SNR. Result: noisy red flowers!"

The author goes on to say that one colleague has measured the relative sensitivity of the sensor in his Canon 5D Mark II to be 5:2:1 in the Green:Blue:Red channels respectively.  These means that before any white balance corrections are applied in the camera (for the displayed JPEG and associated histogram) or in the raw processor, the intensity of the green channel is 5 times that of the red channel.  Translation:  even though the red channel might look like it is clipping after the white balance corrections are applied, the raw data is not clipping, and the detail can be recovered by using the exposure slider or highlight recovery tools in the raw processor.

Why is this understanding useful?  Because for most "normal" lighting conditions (i.e. those taken in light with a color temperature close to "daylight" or flash), we can simply look at the green channel in our camera's RGB histogram to see if the data is clipping.  Since we know that there is a "multiplier" being applied to the red and blue channels in order to correct the white balance, we can (in normal light) assume that if the green channel exposure isn't clipping, then the red and blue channels aren't being clipped in the actual raw data either.

Bottom line:  With current digital cameras, in lighting with "normal" color temperatures (daylight or flash), the green color channel of your RGB histogram is the best tool you can use to "optimize" your exposure.  The green channel histogram should be brought as close to the right side of the scale as possible without clipping.

(To expound further, when the color temperature of your light source is at the warmest end of the range your camera will effectively record (the warmest incandescent colors your camera can correctly display) then the white balance "multiplier" for the red channel will be 1, and the multiplier for the blue channel will be set to the maximum amount in order to neutralize the resulting warm color cast in the image.  The opposite applies to the coolest light sources.  If you are shooting under light that has a color temperature in the 8000 degrees kelvin range (very blue), then the white balance multiplier for the blue channel will be set at 1, and the red channel will be set to the maximum amount the camera or software allows in order to neutralize the blue color cast.  Make sense?  So how is this information useful?  Under extremely warm lighting (incandescent lights) you will have to watch both the green and red channels for clipping, while under extremely cool lighting (some fluorescent bulbs) you will have to watch both the green and blue channels for clipping.)

Here's the article :
 
http://www.luminous-landscape.com/tutorials/optimizing_exposure.shtml

Enjoy!

Keith

[prairiedust]

Good to know!

[girod]

But Keith, what about the pictorial intent (DOF, motion blur)? I think that this is more important or has more priority than capturing the maximal light that the sensor can record. For instance, in the black cat/white cat scenario - my first considerations/concerns would have been a) are they moving? and do I want to freeze them? b) do I want to isolate them or not? - then I would verify if my chosen aperture/shutter speed would clip highlights the detail of which I want rendered, and if I'm using my D7K - I wouldn't worry about shadows. I can render to my taste the degree of blackness/whiteness of the cats in processing but nothing I can do about the DOF/motion blur.

So I think that ETTR is not an optimal (maximal, yes) exposure technique because what good is an image if it has the maximal data but void of one's desired/required DOF and motion blur. If one is more conscious about ETTR, the tendency is to compromise one's pictorial intent. It's like, to what end does one records light?

What do you think?

[keithsnell]

But Keith, what about the pictorial intent (DOF, motion blur)? I think that this is more important or has more priority than capturing the maximal light that the sensor can record.

So I think that ETTR is not an optimal (maximal, yes) exposure technique because what good is an image if it has the maximal data but void of one's desired/required DOF and motion blur. If one is more conscious about ETTR, the tendency is to compromise one's pictorial intent. It's like, to what end does one records light?

What do you think?

I agree 100%, pictoral/artistic intent should be higher priority than simply exposing to the right.  I usually select the aperture, shutter speed and ISO needed to meet my intent, then verify that the green channel isn't clipping (or isn't more than one stop underexposed).  If I need to adjust my shooting parameters to prevent clipping (or significant under exposure) then I will.  You are correct that "pictoral intent" should take priority.

By the way, on Nikon cameras you can select what channel is used as a basis for the "flashing highlights" warning.  the default is the luminosity histogram, but this isn't always a good indication of clipping in the raw data.  The green channel provides a very accurate indicator of clipping in the raw data, so I have my cameras set so that the flashing highlights warning is driven by the green channel.

Keith

[girod]

Thanks Keith.

I didn't know that R or G or B can be selected as the basis for the flashing highlights. Where in the menu can you select this?

[girod]

Hello Keith,

I'm sure that you've read Emil Martinec's explanation on why ETTR works contrary to Michael Reichmann's:

http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/noise-p3.html#ETTR


jaime

[keithsnell]

Thanks Keith.

I didn't know that R or G or B can be selected as the basis for the flashing highlights. Where in the menu can you select this?

Hi Jaime,

You select it on the RGB histogram display itself.  In the lower right of the RGB histogram display screen there is a prompt that indicates that if you select the thumbnail display button (minus zoom button) and move the multi-selector right or left you can select which channel is used for the flashing highlights display.  (pg 166 in the D7000 manual.)

[keithsnell]

Hello Keith,

I'm sure that you've read Emil Martinec's explanation on why ETTR works contrary to Michael Reichmann's:

http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/noise-p3.html#ETTR


jaime

They are both right.  I used to shoot with a digital camera that saved 10-bit raw files.  You could definitely see more "noise" in the shadow areas because of the quantization error.  Now that digital cameras have higher bit depths, it's less noticeable, but still there.

Michael Reichmann said in his original article, and reiterates in this one, that there are several reasons for exposing to the right.  In his first article he said:  "The first is that it will maximize the signal to noise ratio. The second is that it will minimize the posterization and noise that potentially occurs in the darker regions of the image."  

Emil argues that the reason to expose to the right is driven purely by the S/N ratio (which in turn is determined by the read noise).  He says "Read noise at high ISO is much smaller than read noise at low ISO, in terms of the error in photon counting that it represents. Thus, better image quality is obtained for using the highest ISO for which the signal is not clipped."  This statement has been turned on it's head by the read noise characteristics of the D7000, where read noise is essentially equivalent at both low and high ISOs.  HIs statement that "better image quality is obtained for using the highest ISO for which the signal is not clipped," is very dependent on the camera's analog to digital conversion circuitry.  His statement is true for the D3 and D700, but only up to ISO 800, and simply not true for the D7000.  So, Emil slams Michael Reichmann for generalizing, but he is guilty of the same thing himself.

The bottom line is that they both agree that exposing to the right, without clipping, is necessary in order to optimize your raw data capture.  Neither one of them does a sufficient job discussing the fact that sometimes we shouldn't try to prevent clipping at all cost, since trying to preserve specular highlights can result in very significant underexposure for the "important" parts of the image.  It all comes back to the concept that exposing based on your "pictoral intent" is the primary consideration.  That's why the Zone System is a much more effective exposure concept than ETTR.

The challenge in trying to convey any concept is to keep it simple enough to be understood, without oversimplifying so much that your statements become inaccurate.