I think this procedure is flawed in one respect: light at the dark and light ends of the histogram may be captured by the sensor, but it only matters if there is meaningful, interpretable, and visible detail. One sensor may have an extra stop of dynamic range, but if that extra stop is no more than nearly blown out highlights then who cares? That's the main knock on digital sensors versus film.
DrPablo wrote:
I think this procedure is flawed in one respect: light at the dark and light ends of the histogram may be captured by the sensor, but it only matters if there is meaningful, interpretable, and visible detail. One sensor may have an extra stop of dynamic range, but if that extra stop is no more than nearly blown out highlights then who cares? That's the main knock on digital sensors versus film.
Perhaps the overall range is a bit acedemic, but there is often useful information in what first appears at first glance to be blown-out highlights. Shooting in RAW and making curve and level adjustments can really extract some of that detail that one might think is lost.
At the other end of the spectrum, the noise floor is of significant importance as well. I don't think that film can deal with ISO 1600 as well as the newer generations of digital cameras can, and particularly those with large photosites.
DrPablo wrote:
I think this procedure is flawed in one respect: light at the dark and light ends of the histogram may be captured by the sensor, but it only matters if there is meaningful, interpretable, and visible detail. One sensor may have an extra stop of dynamic range, but if that extra stop is no more than nearly blown out highlights then who cares? That's the main knock on digital sensors versus film.
Dr. Pablo,
The bright side can be clearly measured. It is about whether the signal is below or above the sensor range. My procedure can tell clearly.
The lower side, as we have been discussing, is a little fuzzy. One can argue that the image is too noisy. Another can argue that there is some information that he can extract out. My procedure is to check if the signal can be separated from the noise.
The low end measurement can be devised differently. And a stop here or there is not important. What's important is if every one can measure it consistently.
DrPablo wrote:
I think this procedure is flawed in one respect: light at the dark and light ends of the histogram may be captured by the sensor, but it only matters if there is meaningful, interpretable, and visible detail. One sensor may have an extra stop of dynamic range, but if that extra stop is no more than nearly blown out highlights then who cares? That's the main knock on digital sensors versus film.
Film and digital sensors are not directly comparable anyway. Digital is closer to colour transparency in the way that it works, and in this case DR is much better in digital. Comparing against negative film, yes digital highlights don't taper off, they hit the end stops and blow out. At the other end of the range though the noise in shadows is MUCH better. I remember seeing a plot of digital vs film response including noise and, while the digital plot diverged only a little from ideal in the shadows, the film plot got very noisy well before it's ultimate limit.
What it comes down to is you need to meter for highlights rather than shadows. If you allow for that then most digital camera's (usable) DR isn't wildly different from film.
Jack Flesher wrote:
You can also do this with one shot by photographing a standard Kodak trnsmission step-wedge on a good (evenly lit) light-table.
Thanks for the suggestion, Jack. Oh BTW, long time no see
Yes, there are multiple ways to do this. I was trying to come up with procedures that most of us can do consistently without equipment or special stuff.
steve_t wrote:
Film and digital sensors are not directly comparable anyway. Digital is closer to colour transparency in the way that it works, and in this case DR is much better in digital. Comparing against negative film, yes digital highlights don't taper off, they hit the end stops and blow out. At the other end of the range though the noise in shadows is MUCH better. I remember seeing a plot of digital vs film response including noise and, while the digital plot diverged only a little from ideal in the shadows, the film plot got very noisy well before it's ultimate limit.
What it comes down to is you need to meter for highlights rather than shadows. If you allow for that then most digital camera's (usable) DR isn't wildly different from film....Show more →
That may be true, but few would argue that highlight detail is more important than shadow detail. This is a biological phenomenon, not just aesthetic. The photoreceptive cells in our retina are rods and cones. Cones register color and discriminate detail better than rods, but rods are better at picking up light in dim conditions. Our highly detail-sensitive cones just don't work under low light. Rods are sensitive to low light, but not to color; furthermore, they are distributed more diffusely in the periphery of the retina, away from the fovea (the area of highest visual acuity), whereas the fovea is stacked with cones. So our eyes have much more difficulty discriminating detail in low light than in bright light.
This issue extends to photographs; one rarely hears photos criticized for clipped shadows, but people comment about blown highlights all the time. Picture a nice gaussian distribution of pixels over the histogram, and you are given the choice of clipping either 5% of highlights or 5% of shadows. Almost invariably the clipped highlight version will look worse. So even if the absolute dynamic range between film x and digital sensor y are the same, the value of that range (photographic value, not mathematical value) is not necessarily the same -- highlight detail means more in the end than shadow detail.
This is why I look forward, some day in the future, to the HDR-type sensors that include a little ND-filtered sensor right next to the regular RGB sensor, so that image sensors are better at capturing detail in the highlights.
I tried this experiment with a bit different approach.
I was prompted to try the following approach by the discussion that the "test" was essentially looking at SN ratio. My attempt is to look at darkest shadow to brightest highlight detail. That is what we look for in an image.
I lit a paper towel with a side light and then shot a series of frames using Pondria's method. shutter to +2EV then kept slowing the shutter.
Then to -2EV and kept increasing shutter speed.
I then looked at the images in ACR looking for just the hint of "shadow detail" in the under exposed images and just a hint of "highlight detail" in the same area on the over exposed frames. The frame with a hint of highlight detail was 1/20 while the one with a hint of shadow detail was 1/3200. That comes to just a bit more than 7 stops.
As someone pointed out, the real challenge is to get detail in the highlights and shadows for the final "presentation" of the image. Even the best photo printers available today are probably more limiting than the digital camera we are using.
All the talk about "pop" and "wow" in Guy's DMR thread are seem to be oriented toward midtone contrast than a broad dynamic range.
David,
Your method also makes good sense. Basically you look at the 2D image for the "Hint of the shadow/highlight" details while I look at the 1-D histogram for the separation of the bell curve.
Your final number seems to be within the range of others already reported for 10D. So, I am interested in knowing if you had a chance to compare the high and low end differences between your method and my method.
I *guess* your method is more conservative on the high side and more aggressive on the low side.
DrPablo wrote:
That may be true, but few would argue that highlight detail is more important than shadow detail. This is a biological phenomenon, not just aesthetic. The photoreceptive cells in our retina are rods and cones. Cones register color and discriminate detail better than rods, ...
Which ignored the point entirely. If you meter a shot with film so that the shadows are less noisy then the highlights start to curve off but don't blow. This is the way that film photographers have worked for years. With digital you can meter the highlights so they don't blow and not have to worry about shadow noise.
Translating this, with digital it's better to under expose (providing you have low noise and good DR). With film it's better to over expose as the highlights will, for the most part, take care of themselves. You just need to remember the difference between the two and work accordingly.
steve_t wrote:
Translating this, with digital it's better to under expose (providing you have low noise and good DR).
Steve, so you are advocating "expose to the left"? I like to "expose to the right" because I am convinced that it gives the most information. Can you elaborate on the superiority of your thinking? Thanks!
Wiseman wrote:
Steve, so you are advocating "expose to the left"? I like to "expose to the right" because I am convinced that it gives the most information. Can you elaborate on the superiority of your thinking? Thanks!
I think Steve's proposing "Expose to the right". Both of you are saying the same thing. You make sure that the highlight is not blown. And I also agree.
Inherent noise or self noise is Lower by far in a CCD than a CMOS chip. Digic is the savior for CMOS. They are cheaper to design , build and lower power , the achilles heel has been noise , BUT , the savior for it is the design allowing NR to be done at the photosite level . So as in all things in photography , compramises , friends , compramises
OK,
Here is an other version of the test.
I used the same technique Pondria described but added a black wool sock to the paper towel. Then made a series of exposures at F8 ranging from 1/4000 to 2 seconds.
I then looked for first image where highlights could be brought below 255 with the ACR exposure slider and the shadows brought to near 0 with the shadow slider. That is the top section.
I then looked for the image where the highlights could be pulled up to nearly 255 and the shadow at least 0. That is the middle section.
The final section is an image between these two extremes where The highlights could be brought up with the exposure slider and the shadows at or a bit above 0 with exceptable noise in the shadows. That is the bottom section.
What is interesting is that the highlight side of each section looks very similar while the shadow side looks much different.
David, thank you for the interesting testing. It is also interesting that the color tone of the black cloth changes, while the white paper towel maintains the WB well.
10DFT wrote:
It would be an interesting comparison.
I took a look at Imatest.
Quick Note ( You are welcome to add or correct )
1. They use the step chart target. But it is not the fundamental difference. For instance, you can produce one composite step chart by Copy-Paste from the shots from my methods. ( A lot of work though )
2. They use JPEG. The data may be already truncated to 8bit - No big deal, though.
3. At high end, they take the 98% point ( 250/255 ), In my method, I take the max below the saturation.
4. At low end, they define 3 relative noise levels - High ( 0.1 stop ), Med, Low ( 1stop ). As this is completely human-free operation, this is more consistent than my method.
What's the bottom line ?
For 10D, I got 3 reports ( 7, 7 1/3, 8 ).
Imatest 10D example shows ( H-6, M-7.5, L-8 )
So, I think my methods is "approximately" equivalent to their Medium quality criteria measurement.
( Well, to my credit, my method is a lot simpler to do and to understand )
)