I think cameras already have the ability to record more dynamic range than both monitor and paper are able to reproduce. Techniques to enhance dynamic range are really closer to compression so you can actually print and see all the lows, mids and highs (e.g., using 2 or 3 RAW conversions & blending in layers, contrast mask, etc.).
Thanks for the link. It would be great if we could get some data points for the 5Dmk2, the 50D, and the 7D using this procedure. Alternative proceudures are welcomed, too.
Camera processors are still too slow and storage still too small. I can't wait for cameras to start recording in 32bit floating points, gone will the exposure bracketing be... True HDR at a click of a button. Now where's that drooling smiley gone to...?
Gochugogi wrote:
I think cameras already have the ability to record more dynamic range than both monitor and paper are able to reproduce. Techniques to enhance dynamic range are really closer to compression so you can actually print and see all the lows, mids and highs (e.g., using 2 or 3 RAW conversions & blending in layers, contrast mask, etc.).
The critical factor is capturing scene data in a single exposure. Even with film the negative has a greater range than the print paper. When photographers shot exclusively in B&W scene contrast wasn't a problem. The Adams Zone system, which I used for B&W adjusts neg. development time to fit different scene ranges to the same #2 grade paper, which could reproduce a density range of about 10 stops (.30ND x 10).
Color film can't be manipulated like B&W to change contrast because the color recording layers get out of whack creating color shifts in neutrals. The range of color print paper was also shorter than B&W. Because of that photographers had to learn how to use flash effectively to alter the foreground of an outdoor scene so the density range of the neg. matched the shorter range of the print paper.
With digital the physical design of the sensor cells is the factor which defines the dynamic range and they can't record detail in outdoor many outdoor scenes. Exposure ends when the brightest areas in the scene max-out the capacity of the sensor site to record photons. The bigger the "bucket" the longer it takes to fill, giving darker shadows more time to record a signal over the noise. That's way larger sensors sites have better DRs.
A simple empirical way to measure effective DR is to shoot a bracketed exposure of a gray card, exposing it so it is reproduced from 245 (textured white eyedropper reading) to the noise threshold. I did that with my 20D and 24-70mm lens and got these results:
A little over 6 stops of detail is can discern in actual images (245 to 30). The dilemma is that an average outdoor scene will have a 10 stop range, and a beach or snow scene 12 or more. So I keep a flash on the camera when shooting outdoors, and to record a full range of detail in the foreground shoot into the shadows of the ambient so the flash will not overlap ambient highlights and shadows at the same time, expose ambient to keep highlights below clipping, shoot into the shadow of the ambient, then raise the shadow side with flash.
Here's a high speed flash test I did with my 20D: I started with baseline shot exposed for detail in the white towel (highlight exposure benchmark) using the clipping warning on the towel in mostly flat light on the target: http://super.nova.org/TP/DR_FlatLight.jpg
Same target, exposed for the same highlights, shot from the other direction: http://super.nova.org/TP/DR_Backlight.jpg
The overall tonal ranges in both photos are actually the same by the first one looks better exposed perceptually because perception is based largely on expectations. We know the card is middle gray and the towel white and large parts of both are highlighted in the first shot but not in the second. Then I reached up and turned on the flash, set to FEC =0:
When the ambient is kept below clipping the evaluative ETTL-II metering works quite amazingly to fill the "hole" in the ambient pattern created by the foreground object. The foreground might even look slightly overexposed perceptually because given the context we expect the shaded side to be slightly darker, but in the RAW file there is still a detail in the sunny parts of the white towel with the shaded side a slightly darker shade of white. Checking just the foreground in Levels I found that the range of tones on it from black to "paper" white fit the sensor exactly, which is the definition of correct exposure in the technical sense.
Perceptually the background still looks darker than normal (seen by eye) due to the fact I exposed for the ambient highlights and the sensor range can't handle the scene contrast.
In a situation like a portrait the perceptual effect of the darker background could be controlled by cropping and showing only what is require for good compositional balance. That's pretty how photographers learned to cope when scene contrast exceed film or sensor. They would expose to get the perceptually critical face correctly exposed at the expense of blowing some highlights like the sky and losing some shadow detail, then mitigate the damage with cropping so its wasn't noticed.
That's why defining "correct" exposure is so difficult and contentious. Technically correct exposure requires detail over the entire tonal scale, as in this studio lit shot:
Fitting scene to sensor in the studio is trivial. Start with a dark room. Turn on fill over the camera where it light everything the camera sees evenly. Raise it until shadow detail is seen in the darkest areas where detail is desired:
Once the shadows are raised above noise for that given ISO/aperture/shutter speed creating 3D modeling is just a matter of adding off axis key and accent backlight until the clipping is seen, then backing off intensity a bit. http://super.nova.org/TP/LE04.jpg
Its ridiculously simple in the studio to perfectly match scene to sensor with just the camera feedback and your eyeballs. Its nearly as easy with flash outdoors as illustrated above if there is a highlight reference value in the scene in both the ambient and flash lit parts of the photo.
So rather than fretting over the limits of DR in cameras, which I don't think will increase much without breakthrough in sensor technology, I suggest photographers learn how to use flash effectively indoors and out when shooting on a tripod with HDR isn't an option. Flash will not work in all situations, but I've found many ways to use it effectively
I did some quick testing (will post the results later), but it looks like at a quick glance that my 7D copy at ISO 100 has between 7 and 8 stops of DR.
I would like for the camera electronics to deliver all the pixel DR that Canon sensors currently seem to have, which is over 14 stops for the 1D3 and may approach that for the 7D. The sensor DR is the electron count at base ISO RAW saturation, divided by the read noise at high ISO (typically ISO 1600 or above). If the electronics downstream of the sensor in the signal processing chain didn't get in the way, there would be no reason to worry about ISO -- just set the desired shutter speed and aperture and you're good to go. But since the downstream electronics is quite noisy, the delivered DR is less than 12 stops, and one is then forced to choose which part of the sensor DR to keep, by dialing the ISO amplification (the lower part of the sensor DR for high ISO, the upper part for low ISO). Having cleaner ISO amplification and ADC would go a long way to addressing DR issues in current Canon DSLR's.
I will rerun Pondria's DR protocol for my 1DIII and do it for my 5D II at the same time.
Unfortunately, it may be two or three weeks before I can do this controlled comparison.
Maybe Canon would like to hire Emil Martin as an expert consultant for product planning!
Pondria wrote:
Cameron12x,
Is it 7 1/3 or 7 2/3 ? I'm gonna add it to the data base.
7 2/3, but I'm going to retest.
ejmartin wrote:
I would like for the camera electronics to deliver all the pixel DR that Canon sensors currently seem to have, which is over 14 stops for the 1D3 and may approach that for the 7D. The sensor DR is the electron count at base ISO RAW saturation, divided by the read noise at high ISO (typically ISO 1600 or above). If the electronics downstream of the sensor in the signal processing chain didn't get in the way, there would be no reason to worry about ISO -- just set the desired shutter speed and aperture and you're good to go. But since the downstream electronics is quite noisy, the delivered DR is less than 12 stops, and one is then forced to choose which part of the sensor DR to keep, by dialing the ISO amplification (the lower part of the sensor DR for high ISO, the upper part for low ISO). Having cleaner ISO amplification and ADC would go a long way to addressing DR issues in current Canon DSLR's....Show more →
Do we potentially lose (6) or more stops with the downstream electronics based on the testing protocol suggested earlier in this thread?
Can you suggest an alternative way to field-test DR?
Many of us would probably like to better understand this. Can you explain in more detail how the theoretical limit of 14 stops was derived? Thanks!
cameron12x wrote:
Do we potentially lose (6) or more stops with the downstream electronics based on the testing protocol suggested earlier in this thread?
Can you suggest an alternative way to field-test DR?
Many of us would probably like to better understand this. Can you explain in more detail how the theoretical limit of 14 stops was derived? Thanks!
I think the procedure that I propose is probably the easiest way to get the DR consistently.
We talked a lot about "The 14bit-ness" 2 years ago when 1D3 came out. New comers, there is really no new issue that we didn't talk about
If you are passionate about this, you may want to read through the whole thread. https://www.fredmiranda.com/forum/topic/543552/6&year=2007#4755375
