Anti-HDR (Composing for Dynamic Range), Photography Assignment for 13 - 19 June

[keithsnell]

You can blame this assignment topic on Marilyn and this image:

Through the Tunnel
Photographed by Marilyn McKinney

Why?  Because here's my shot of the same location:

YIKES!  I'm ashamed to admit that's my image.  So how did I blow it so badly?  The answer is that I visualized the shot using a wide angle lens in order to accentuate the distance down to the bottom of the ravine, but I failed to anticipate the high contrast of the surrounding scene.  It would have been much better if, like Marilyn, I had composed my image to eliminate the high contrast areas in the scene.

Here's another example.  First my shot:

Again, I visualized the composition using a wide angle lens to accentuate the length of the train, but I blew it by including high contrast areas in the scene that can't be recovered in post processing.  I should have recognized that the dynamic range of the scene was greater than the sensor could handle and recomposed my image accordingly.  Here's Rebecca's shot, taken just a couple of seconds earlier than mine:

White Pass Railway
Photographed by Rebecca

She zoomed in on the most interesting part of the scene, and eliminated the high contrast areas in the process.

Reviewing my images from the trip, and seeing the many shots that just didn't make the cut due to the high contrast, reminded me that I should be paying more attention to the dynamic range of the scene when I'm composing my images.  In my film days, when I primarily shot transparency film (because that's what the magazines accepted for publication), I learned very quickly that most transparency films could not handle high contrast scenes.  The majority could only handle a range of around 5 stops, with +2.5 being rendered as "pure white" and -2.5 being rendered as deep black.  This high contrast rendering added more "punch" to the images, but limited the photographer to only photographing scenes that would fit within the dynamic range of the slide film (or accepting that some parts of the image would be recorded as deep shadows).  

Unfortunately, with the advent of digital, many people fell into the trap of believing that they could effectively capture and render a much higher dynamic range scene.  While there are some exceptions, for the most part this "myth" just isn't true.  Camera manufacturers have "tweaked" the tone curves compared to the early days of digital in order to provide some "roll off" of the highlights and extend the dynamic range that can be rendered, but the reality is that there are limitations to the amount of dynamic range that can be effectively rendered.  

A clue to one of the primary limitations comes from the zone system, where Zone VII is defined as "the brightest area that retains visible detail" and Zone III is defined as the darkest area that retains visible detail.  These definitions are based both on the medium used to render the image (print) AND the limitations of human visual perception.  We can only perceive "good detail" in a narrow range of brightness values, and that range corresponds to a span from about -2 EV below the defined "mid-tone" value to +2 EV above the mid-tone value.  

Try this experiment for yourself.  Put your camera in spot meter and manual exposure and take a series of images zoomed in on a white towel (so that you can later assess the apparent level of detail).  Vary the exposure in 1-stop increments from +3 EV all the way down to -5 EV, then review these images on the camera's LCD in normal room light.  At what point do you stop perceiving "good detail" in the weave of the white towel?  For most people, the range they can perceive "good detail" will stretch from -3 EV to +2 EV.  Sure, you can perceive some detail at slightly lower or higher exposure values, but it is arguable whether or not this can be defined as "good detail."  Below -3 EV and above +2 EV our eyes just aren't as effective at distinguishing the changes in contrast necessary to perceive detail.  

"Changes in contrast" is the key phrase.  Sure, we can "bump up the shadows" or "recover" the highlights to squeeze more dynamic range into an image, but when we do we are compressing a wider dynamic range into a narrower tonal area, thereby reducing the contrast differences for all of the information compressed within that range (and reducing the perception of detail, which is heavily influenced by the "micro contrast" in those areas of the image).  So there's no free lunch.  We can try to "compress the highlights" to recover detail in the brightest whites, but in doing so we're actually reducing the detail (micro-contrast) in the mid-whites.  

There's still some discrepancy between my statement that "we can only perceive good detail in a range from about -2EV below the defined mid-tone value" (which is consistent with the definition of Zone III) and the results of the test I suggested you perform, where you could perceive good detail in the images recorded at -3 EV.  Why the discrepancy?  Because camera manufactures "bump" the tone curve so that the recorded mid-tones are rendered brighter than mid-tone.  This compresses the highlights and "stretches" the shadows, so that in effect an exposure of -3 EV is rendered brighter than if the tone curve were linear, putting it closer to the brightness range where we expect to be able to perceive detail.  

Yes, you can try to capture and render a high dynamic range scene by boosting the shadows, compressing the highlights, or using "tone mapping" to redistribute the tonal values so that they fit within the range of your output device, but in most cases you will be accepting a significant compromise in the rendered tonalities and detail, and will be giving up a "life-like" rendering for one that is more "artistic" in nature.  In many cases, it is much better to consider the dynamic range of the scene and adjust your composition in order to ensure that the tonal values fall within a range that can be effectively rendered.  In most cases, metering the "brightest areas that will retain visible detail" at +2 EV and the "deepest shadow areas" as -4 EV will result in an optimum rendering.

The assignment for the week of 13 - 19 June 2011 is "Anti-HDR (Composing for Dynamic Range)."  You should attempt to find a high dynamic range scene, and then adjust your composition to include only areas within the scene that can be effectively captured and rendered with a "normal" contrast curve.  Please upload your images to the Anti-HDR album in the Weekly Assignments category of the gallery no-later-than midnight, Mountain time (GMT -07:00) on Sunday, 19 June 2011.

I'll look forward to seeing your images.  I'm in remedial training to make myself pay more attention to the dynamic range of a scene and adjust my composition accordingly, so I know this exercise will help me improve my photography.

Keith

[keithsnell]

I'm not sure that I did a very good with my earlier explanation, so I thought the following "test" image might help you visualize what I am talking about.  In the shot below, I used spot metering and manual exposure to expose the white towel at +2 EV on the exposure scale.  This places the towel in Zone VII, defined as the brightest area that retains visible detail.  The metered exposure range in this scene (determined by metering on the white towel and then metering on the black cloth) is 5.3 EV.  In this scene the histogram is extended just as far to the left and right as it could be, without clipping any of the data.  This may seem inconsistent with camera reviews by DPReview and other sites that measure the camera's dynamic range as 9.2 EV.  As we talked about before, the way DPreview performs theses tests is intended to help characterize the sensor performance.  It is NOT intended to be used for an exposure guide.  I can guarantee that if you meter your whites at +4 EV or meter your blacks at -5 EV they will be clipped.  You have to remember that the meter reading (for an area with texture) is measuring (and averaging) a brightness range, as show in the "peaks" in the histogram below.  The "median" of this range must be significantly inside the dynamic range limits of your sensor if you expect to retain all of the data without clipping.

In the image below (and judging by the histogram above) you can see that any more than +2 EV exposure on the white towel would begin to clip the highlights and result in a loss of detail.  (Actually, I had to bring the exposure down 1/6 of a stop in Capture NX2 to prevent clipping the brightest whites.)  Perceptually you should be able to see that in the brightest areas of the towel, your eye is already beginning struggle to see the detail.  (It's there, buy our eyes can't distinguish it well when it is compressed into the highest ranges of the histogram.  You should also be able to see that it is difficult to see very much detail in the texture of the black cloth.  (I should have use a lint brush to remove the lint and dust from the black cloth, since the fine dust and lint can be perceived as detail in the cloth.)

White towel exposed at +2 EV (to place it in Zone VII)

The next image is based on the same raw file, however I used the "shadow protection" slider (set to maximum) and a levels adjustment to bring the black cloth up to a "mid-tone" brightness.  I'm still not clipping any of the whites.  All the data is still there and the right edge of the histogram hasn't moved; however, the mid-tones and whites have all been compressed to the right.  This compression reduces the micro-contrast in these tonal areas, and makes it more difficult to distinguish detail.  Notice however how it is much easier to distinguish detail in the dark cloth when it is rendered as a mid-tone.  (It's also worth noting that the use of the "shadow protection" slider in Capture NX2 provides the "best case" results, minimizing the compression in the highlights.  Other methods (LCH, levels or curves adjustments, or exposure adjustments used in conjunction with highlight protection) will destroy the detail in the highlights much quicker than the "shadow protection" function, which attempts to bump the shadows and mid-tones without affecting the highlights.)

Blacks brought up to mid-tone (as in recovering a face from the shadows)

Histogram of image with blacks brought up to mid-tone.  Whites aren't clipped, but are compressed to the right enough that detail is hard to perceive.

It's also worth noting that this was not a "challenging" scene.  Neither the highlights or shadows were clipped in the "standard" rendering.  With a more challenging scene (normal sunlight with a range of highlights and shadows) recovering clipped highlights and/or shadows could cause much more of a loss in detail or color fidelity than in the example I showed above.  As I said in the first part of this assignment "Yes, you can try to capture and render a high dynamic range scene by boosting the shadows, compressing the highlights, or using "tone mapping" to redistribute the tonal values so that they fit within the range of your output device, but in most cases you will be accepting a significant compromise in the rendered tonalities and detail, and will be giving up a life-like rendering for one that is more "representative."  

If the metered dynamic range of your scene is within 5 to 6 stops, you will probably be able to render the scene realistically.  If the metered dynamic range is more than that, you will be forced to accept compromises in tonality and detail rendering in your output.  Your only other option is to recompose the scene to eliminate high-contrast areas (or use flash to fill in the shadows).

[girod]

Question Keith: What if (with the D7K) you spotmetered the white towel to +3.7 and in postprocessing apply a negative exposure compensation equivalent to -1.7EV, can the white towel be rendered to Zone VII just like as if it has been metered at +2EV?

[keithsnell]

Question Keith: What if (with the D7K) you spotmetered the white towel to +3.7 and in postprocessing apply a negative exposure compensation equivalent to -1.7EV, can the white towel be rendered to Zone VII just like as if it has been metered at +2EV?

Hi Jaime,

Thank you for asking.  The short answer is "no."

Here's an image of that "oh so captivating" white towel spot metered at +2 EV:  (NOTE:  I had to lower the exposure -0.5 EV in Capture NX2 in order to protect the brightest areas of the towel from clipping.)

Spot metered at +2.EV, exposure lowered 0.5 EV in post

Here is exactly the same scene, spot metered at +3.7 EV and then highlight protection set to maximum and -2.0 EV (the maximum available) set on the exposure slider:

spot metered at +3.7 EV, highlight protection set to maximum and -2.0 EV in post

I wouldn't use the +3.7 EV exposure for an area that contained "important detail," i.e., if I intended to place that area in Zone VII in my rendered image.

Keith

[WriteHeart]

I guess I don't really know what the challenge is. Are we just supposed to compose the shot to not include the bright sun or the dark shade? So, if we have a pic say under a tree at noon, then we are to take the pic so the subject is only in the shade and compose for that and not include the bright sun around the tree top? Just trying to figure it out.

[keithsnell]

I guess I don't really know what the challenge is. Are we just supposed to compose the shot to not include the bright sun or the dark shade? So, if we have a pic say under a tree at noon, then we are to take the pic so the subject is only in the shade and compose for that and not include the bright sun around the tree top? Just trying to figure it out.

Yes, that's basically it.  You should compose your image to eliminate any areas that would be too light or too dark to print or display realistically in the final image.

Does that help?

[girod]

Thank you very much Keith for answering my question by demonstrating it. What is then the saturation point of the D7K sensor based on its in-camera exposure meter?

As for the assignment, if I understood it right, is: Composing for Dynamic Range "of 5-6 stops for sRGB JPEG output"? Now I understand why you call it "Anti-HDR". I guess, my entry does not satisfy this requirement. 

[keithsnell]

Thank you very much Keith for answering my question by demonstrating it. What is then the saturation point of the D7K sensor based on its in-camera exposure meter?

Hi Jaime,

I wish that Rawnalyze worked with D7000 files so that I could give you a more precise answer.  Unfortunately the author of that software passed away awhile back, and it hasn't been updated for the new cameras.  I haven't found a substitute yet, or any reputable reference that gives the saturation point of the D7000 based on its in-camera exposure meter.

Based on my experience with the camera (not on any rigorous testing) I would guess the saturation point is just above +3.0 EV.  (The saturation point is somewhere above +3.0 EV, but below +3.7 EV.  It would take further testing on my part to nail that down.)

I typically don't expose for saturation point though, since that doesn't give any margin at all.  If I had exposed the white towel at +3.0 EV instead of +2.0 EV, I would have still had clipping in the brightest parts of the towel.  (Remember that the meter is averaging the brightness of the area under the 3.5mm spot (2.5% of the frame) the spot metering covers.  Some of that area will be brighter than average, so metering right at the saturation point will result in about 50% of the data under that spot being OVER the saturation point.  (The distribution is normally a bell curve, so 50% of the data would be over the mean (or average value).  Make sense?)

As for the assignment, if I understood it right, is: Composing for Dynamic Range "of 5-6 stops for sRGB JPEG output"? Now I understand why you call it "Anti-HDR". I guess, my entry does not satisfy this requirement.  

I fear that I didn't do a very good job describing the assignment.  Yes, "composing for Dynamic Range of 5-6 stops for sRGB JPEG output" is an accurate description of the assignment.  I'm actually glad you submitted your image, since it started me thinking more about how to better describe the concept.

Thank you again for asking questions.  Answering your questions gives me another chance to try to explain the concept a little better.

Keith

[keithsnell]

Here's one more set of examples from images shot this week.  We photographed a church service in a church that has fluorescent "rim" lighting around the back of the altar.  Not only does it present problems with white balance, but it also blows out the background in almost every shot you try to photograph from the pews.

The solution?  Change your viewpoint and recompose to eliminate the bright background.  (Move to the balcony and use a long lens, in this case the 200-400mm f4.) 

D3 and 200-400mm at 250mm and f5.6, 1/60 sec, ISO 6400

The 200-400 is becoming one of my favorite lenses for photographing wedding ceremonies and church services.  You can shoot very unobtrusively from the balcony, and usually get much better shots than you could from the floor.