It'sCanon
 

Ideas to park in your photo-brain

2. Dynamic Range

How Black is Black?

Dynamic range is the measure of an imaging system's brightest subject to darkest subject. It answers the question, "How much tonality can I capture in a single exposure?" and the answer is far from simple.

Digital photography, as it is interpreted by JPEG images, attempts to hang on to a useful dynamic range right up to the point of whitest white. One can see this in an editing program such as Photoshop as a brightness of 100% or and RGB value of 255, 255, 255.

Everything under that point is represented by a smaller percentage or numerical point in RGB space.

In ergonomic terms this means more to us humans when it is represented in percent, so this article will discuss it that way.

At 100% a pure white result prevents there being any brighter detail. The ceiling has been banged into. Your print paper will accumulate no ink; your computer monitor won't push more electrons or your flat screen LCD elements will be completely illuminated.

Presumably, at 0% the image will report nothing but totally black pixels, allowing no more detail to be derived from the exposure. Prints at this point will have piled on so much ink that nothing darker could be added to the paper from the printer.

In the photographic world of film, digital images are like transparencies --slide film-- in that once you've overexposed, that image is going to resist being effectively diminished in brightness during a rescue operation. It's whites are gone, baby, and there ain't no getting them back. Slightly underexposed slides can be printed a bit lighter by simply allowing more light through the enlarger. The contrast of the reversal paper is high enough to tolerate a reasonable amount of exposure increase. Scanning slides into the digital realm also lets the photographer lift and shape the tonalities of the image right up to the point at which highlights are bleached to white.

Negative film presents and opposite picture --pun intended-- in that very bright highlights cause the negative to accumulate a darker image during processing. That darker portion of the image can be teased out to reveal more detail by pumping more light through the negative during printing. An overexposed negative can be often "fixed" by aggressive printing and during a transfer into the digital realm it can be captured with a longer exposure. Later, in the computer, the overexposed highlight realm can be stretched into a more presentable look and the image can often be recovered quite well.

Underexposures in a color negative are problematic. Where shadow detail drops to near --but not quite-- zero, pumping less light through the enlarger won't help. It will just make the darkest part of the print grayer, not more alive with detail. In a computer, the limit for detail rescue has to do with how reliably the scanner has been able to differentiate one micro-tone from another.

Scanners with a longer scale of original digital divisions stand a better chance of pulling out more potential detail at the extremes, but simply having a "24-bit" versus a "48-bit" scanner doesn't make the image automatically better. The questions here would be, "So, what does XYZ-Corp's scanner DO with all that info? Do they let me access the portion of the range that contains useful numbers better than GHI-Corp's scanner?" Look to experienced experts for a sense of that.

In the strictly digital camera realm, people want to know how their digicam performs in terms of dynamic range both from JPEG and RAW images. RAW images are seductively longer in scale than JPEG shots, but just claiming this or that performance advantage or practical advantage rapidly deteriorates into a string of apocryphal stories.

Boundaries

The first limitation digital photos encounter are the span of tonalities between "too bright to see anything brighter" and "too dark to see anything darker." In between these extremes lies our digital image experience. But that raises a secondary issue: OUR digital experience is our eye experience. And its experience is all tied to our retina, viewing conditions, the media being viewed, current dark adaptation, vision issues and even age of the viewer.

Working backwards, it probably comes as no shock to folks to hear that younger eyes see better. Similarly, healthy people often see better than folks with health issues. People who are adapted to the current viewing conditions see better than folks walking in from darker or brighter environments and prints viewed in marginal light won't look as good as those in ideal light.

Computer screens can --and too often are-- be pushed to brightness levels inappropriate for local conditions. We've all seen the family group watching TV in the dark. And forever, TV manufacturers have been warning that this is very stressful on the eyes. People still do it, but these days a new factor has emerged --the Internet surfer staring at their monitor in a dark den. And you wonder where all those little headaches are coming from? But I digress...

Photographic prints are ink, dye or pigment on a page of limited maximum brightness. The blackest black to the whitest white is an easy thing to measure. Your digital camera will serve as a light meter. Since the contrast of an image goes up with reduced viewing surface diffusion, glossy surfaces show the greatest contrast. Matte paper (letter paper, matte surface photo paper) won't get as dead black as glossy paper, so do this test with a glossy image. Print a black spot on a white page. The image isn't thrilling but it will tell you something.

In Photoshop, make a square of black floating above pure white. Print it out on glossy paper. Something like this:

       
 

Shoot this print with your camera in 1/3 stop increments from the exposure that sees the white as bleached, pure white (your histogram will help you spot this starting point --it's the histograph in which the white spike is smack against, but not absorbed into, the right side of the graph.

Back off this exposure a stop or so, just for tolerance sake, and start shooting a string of exposures that are brighter and brighter until the entire image is pure white. The histogram spike of the black square will have progressively marched across the graph until it, too, is absorbed into the white edge of the histograph.

Now count the stops from the frame that portrays white as 100% (RGB 255, 255, 255) to the frame that portrayed black as 100%. One frame back from the "black is white" frame, some tiny evidence of black should be either visible with your eye, or technically visible by slinging around the anchor points of the Levels or Curves controls in Photoshop. If you have an image that still contains some tiny remnant of the black square, you can make this detail visible by opening Levels or Curves and dragging the black point over to the side of the graph where the white point lives. In Curves, this means dragging the lower left anchor all the way straight to the right (assuming you've kept the lower left as the black/black corner as it is, below).

This has the effect of reducing the tonal range of the image to 4 units out of 255. (If you look closely, you can see were the "dotted line" from the darkened anchor angles severely up to the open anchor in the upper right. Those extra pixels on the reference hatch are all that's left of the line.) Any pixels that are 254 instead of 255 will now be significantly darkened to a detectable degree.

So, if this frame is the one that exposes black ink, dye or pigment as white, and the other exposure that shows white as 100% tonality for the first time, the measure of stops between them is the dynamic range of ink on the page.

What do you get? I get 4-1/3 to 5 stops. Not a huge range. Five stops (call it) is only a ratio of 2^5, or 32:1. Can this possibly be SO!!???!!

The 32:1 Club

Welcome to the real world in which black is really 1/32nd of white. At least as far as paint, ink, pigment or dye can produce. If you have a blacker black paint or a whiter white paint, I'm really interested to hear about it. The simple phenomenon we have been living with all these years is that our literal portrayal of paintings, posters, house paint, photographic prints, billboards, traffic signs, and most man-made objects are colored with a range of brightnesses that is only 32:1.

Heresy, you say? Well, of course this assumes that the measure is made in the exact same light so the portrayal of subject matter in shade, low illumination or colored light is not accounted for. And certainly there are loads of things in our world that are whiter than white paper, paint or substances. How about highlights, glossy reflections of lighting? Surfaces in our visual field that are getting many times more light than other stuff? Glints? Reflections off chrome?

That's where our eyes shine, so to speak. The living film in your head can sort out a dynamic range of around 30,000:1 on a good day so that piddling little 32:1 dynamic range is a piece of cake / easy as pie /walk in the park / no brainer.

At this point I would like to extend a warm gesture of thanks to the makers of vision. No matter what your creed, philosophy or science, you gotta appreciate that vision exists at all. Without it, digital photography would be meaningless and we would all have to find other things to perk our interest. Something strictly tactile. Basket weaving, perhaps.

The 1000:1 Club

Digital cameras do something your eyes do; they see the world of objects in various lighting situations. Shade, shadow, dimmed lighting instruments and various colors of light are all captured with your digital camera to more or less degrees of fidelity, so it should not be surprising to learn that digital cameras have a MUCH wider range of dynamics than the prints they produce. The print's a physical object while the world is a wide dynamic place full of subject matter in a huge range of conditions.

A similar test to the one that revealed the range of blackest black can be used to reveal the total dynamic limits of a digital camera. The question now becomes, "What is the range from the whitest bleached white to the point at which all shadow detail is lost completely in blackest black?"

The same graphic will help us find this point, too.

As exposures drop lower and lower, the whole image gets darker. Starting at the point that shows the white paper as first banging its head against the white ceiling, we shoot progressively darker shots until the f-stop is so small and exposure time is so brief that the image of a black square against the white paper can no longer support any semblance of definition.

That's easy to say, but remember that digital images have some foo factors inherent in them. A foo factor, you may remember from Logic Class is something that makes you say, "Aw, foo!" It's some unexpected reality that screws up a nice, neat theory or method of testing something. An example of a digital image foo factor is this: No two photosites are exactly alike. Every one of those tiny light reading circuits is inherently built to some tolerance. They're close, but some of them are further out of spec than others.

To our eyes, this phenomenon shows up as "noise" or "grain." The same thing happens in film and even your eyes. (Try to observe the random tonalities in shadows as you are dark-adapted in a barely illuminated situation. Some of that is photon-noise caused by statistically fewer photons reaching your retinal rod cells, and some of it is the noise of the visual system of eye/brain.)

To make an image that for sure has no detail, put the lens cap on the lens and shoot a very brief exposure. This would define the absolute bottom of the dynamic range. No matter what you did with Curves to this sort of image, all you would get would be evenly distributed noise.

To test a dark image for tiny residual image capability, do the opposite of the highlight test. Move the upper right box in Curves to the extreme left, amplifying all pixels with values 1, 2, 3 and 4. If you can still see where the black square was, you haven't gone to the maximum limit. But, my gosh, look at all that noise! Is it practical to insist that the detectable dynamic range of an imaging system includes every speck of deep, noisy, grungy shadow detail?

Technically, yes and no. Dynamic range in an absolute sense would include the farthest limits, but as Leonardo Da Vinci once said, "You have to draw the line somewhere." There's a practical limit to how black is black to qualify as black black.

There are ways of technically defining that point, but you must know by now that I prefer human-scale, practical ways of measuring things. Why take some mathematician's word for something? Is he a photographer? Does he care about image or does he care about math?

At some point in the process of gathering darker and darker images, you will be collecting shots that will look totally black on your computer monitor and print as total black. You could put black lettering over the image and no print in the world would let you read the text. Testing this in Photoshop is duck soup. Text will successfully be lost in the random noise of the image--up to a point.

Tip: Use big solid letters.

You might recover something using the Curves test, above, but don't call that an image unless it passes the MCT test. MCT = Mom can tell. If an observer can't make out the lettering, even after seeing what it looks like, then that level of exposure is equal to White = Black, the definition of the bottom of any valid tonal dynamics and the base of the images effective dynamic range.

Here are some helpful graphics.

On which of the following do you see text over the black? It's there on all of them, but your monitor's exact setting will likely hide it from view for one or more of them:

1

2

3

4

I really can't speak for your computer monitor's state of tweak, but if you don't see the word TEST on the #4 panel, you need to toast up your Internet viewing conditions. See this page for some help.

My own monitor can just barely eke out a visible text overlay on image #3 which is one full stop darker than the one underlying image #4. Image #2 and #1 look black to me, but perhaps your computer monitor or LCD panel does reveal them. If #1 is very visible, I would suspect that your display is way cranked up, or has some viewing issues.

So in my own dynamic test, I'm using #3 as my black base image. My white base image is 9-2/3 stops higher in exposure as seen from these images.

Actual test shots showed a tiny improvement over these figures. The panels above are one stop apart, but exposures made at 1/3-stop increments gave a tad more readability. Conclusion: Call it a solid 10 stops on a Canon 20D shooting Incandescent at ISO 100.

Ten Stops

I'm not surprised to hear that a digital camera can grab a JPEG image with ten stops of truly differentiated dynamic range. Repairing underexposed images relies on the factoid that it's easier to lift low values in an image than to lower bright values.

Here's a chip chart made out of actual image chunks from shots made one stop apart:

It's easy to see how one-stop increments progress from bleached white on the right to maximum black on the left. My monitor shows easily the boundary between squares right up to the last two dark ones. Those look run together, but lightening the whole chip set using the Curves technique above does reveal them as well differentiated:

I guess that purists will call that eleventh square another step in dynamic range, and technically, they'd be right, but if it won't print as a deeper black and 99+% of all monitors won't see it as an obvious zone of extra shadow detail, then it has passed out of the realm of the practical and into the esoteric realm of the scientifically exact. I don't think people buy Ansel Adams prints because they're scientifically precise, though.

So...

While this information doesn't tell you anything about RAW images, it brings a little known and often poorly understood aspect of digital photography into more tangible focus: Digital dynamic range with some of the more current imaging systems hangs on to around ten stops of brightness. That's about 1000:1 in layman's terms and is NOT a bad range of image dynamics at all.

Further, the tests done here can be done on your own camera. No need to take that mathematician's word for it, eh?

Plus, it includes a practical way of defining the upper and lower limits. E.g. upper is the point at which white reaches maximum RGB 255, 255, 255, and lower is the point at which a white subject becomes so severely underexposed that it can't reliably be differentiated from pure computer black text with an RGB value of 0, 0, 0.

Like any test procedure, more questions pile up on the other side. We've tackled one question, but what happens when small changes are introduced into the process?

If this had been shot with Daylight WB under direct sun, would the dynamic range improve? How much?

If the camera's Contrast adjustment were lowered, what effect would that have?

How does this apply to other cameras?

Did that #1 image up there really have TEST written all over it? Well, at least we can answer that one for sure. Here it is with its Curves crunched to the ceiling:

-iNova


Get the eBooks

For your Canon Digital Rebel- DSLR: Canon Digital Rebel EOS 300D eBook is already on the shelf. DSLR: Canon Digital Rebel XT is next. Then DSLR: Canon 20D follows. These are great cameras with a great deal inside them, so we put a great deal of information into their eBooks. You'll find things here that nobody else ever told you about and not even Canon knew or could discuss.

The existing iNova eBooks, DSLR: Nikon D70 has become the most popular premium eBook of all time. Mastering Nikon Compact Digital Cameras and The SONY Advanced Cyber-shot eBook are also available. Click on the image for the appropriate order page.

 
--Peter iNova (peter.inova@mac.com)

itscanon is © 2006, Peter iNova. All rights reserved. Images by the author. They're © 2006, too.
Note: This is a private series of page not supported by or associated with Canon Inc.