Thom Hogan's interesting points regarding DxOMark scoring and the future

[Ellis Vener]

it's worth a read

http://www.dslrbodies.com/newsviews/do-you-believe-in-dxomark.html

At the end he makes some really interesting comments, ones I happen to agree with

"I guess to end this discussion, I should probably suggest what we Nikon users want to see on the DxOMark tests for the next Nikon. We’ve got it pretty good at the moment, after all. Incremental movement of the dynamic range chart upward would always be welcome, but I think we’re now very close to the point where we want to see something entirely different. We really want to see base dynamic range unbound (e.g. the two-shot type DR that the Apple iPhone 6 is doing, and which Sony’s latest patents all hint at). Beyond that, we want to see flattening of the dynamic range slope as ISO is boosted (less of a hit per doubling of ISO).

Simply put, we’re deep ending into small, incremental gains that aren’t pushing us very much further above the “good enough” bar. What we need is to push that bar far forward again, and that’s going to take another sensor generation/design, I think. Until then, you can all anguish over small differences in DxOMark measurements, but I’d be surprised that those show up anywhere near as clearly in your images. "

[dwswager]

it's worth a read

http://www.dslrbodies.com/newsviews/do-you-believe-in-dxomark.html

At the end he makes some really interesting comments, ones I happen to agree with

"I guess to end this discussion, I should probably suggest what we Nikon users want to see on the DxOMark tests for the next Nikon. We’ve got it pretty good at the moment, after all. Incremental movement of the dynamic range chart upward would always be welcome, but I think we’re now very close to the point where we want to see something entirely different. We really want to see base dynamic range unbound (e.g. the two-shot type DR that the Apple iPhone 6 is doing, and which Sony’s latest patents all hint at). Beyond that, we want to see flattening of the dynamic range slope as ISO is boosted (less of a hit per doubling of ISO).

Simply put, we’re deep ending into small, incremental gains that aren’t pushing us very much further above the “good enough” bar. What we need is to push that bar far forward again, and that’s going to take another sensor generation/design, I think. Until then, you can all anguish over small differences in DxOMark measurements, but I’d be surprised that those show up anywhere near as clearly in your images. "

I think he hits it pretty good.  The 7DmkII does pretty well at higher ISOs, but in the end, the $800 cheaper D7100 is a better bet from a image quality standpoint.  He's a little more charitable concerning the progression from 7D to 7DmkII than I would have been.  I would expect to see more than just a stop of high ISO performance out of a replacement for an older camera.  And considering it is a year or more newer than the D7100 it should have been more competitive with the $850 cheaper D7100.  Obviously, Canon is attempting to trade features and speed for their inability to deliver a better sensor.  It's up to the purchaser to determine if that is a good trade for them.

I also agree with his final assessment of incremental gains.  I hope that pushes all the makers to more usability and extensibility as a competitive area.

[Glenn NK]

Obviously, Canon is attempting to trade features and speed for their inability to deliver a better sensor.

Perhaps what's also obvious is that there's more to a camera than the sensor.

Particularly for an action shooter.

[synn]

Perhaps what's also obvious is that there's more to a camera than the sensor.

Particularly for an action shooter.

Considering that Thom is a very accomplished wildlife shooter, I don't think he has ignored that point.

[Glenn NK]

Check to see what BIF guys say on Naturescapes.

[shadowblade]

By definition, DR has to go down by 1 stop for each stop of increased ISO. If it doesn't, it's either not a true 1-stop increase in ISO (stated and measured ISO can differ by a huge amount) or the sensor is being polluted by a large amount of read noise (like with Canon sensors at low ISO).

The way to increase DR at high ISO is to improve the capacity of the photon wells, thereby increasing DR at low ISO (which flows through to high ISO).

[Jim Kasson]

Beyond that, we want to see flattening of the dynamic range slope as ISO is boosted (less of a hit per doubling of ISO).

If your metric of dynamic range is the ratio of full-scale to the average signal level that produces a signal-to-noise ratio of around 10 or so, which happens to be mine, without a radical change to the design of the sensor, there's not much progress to be made here. If he's talking about engineering dynamic range, which is read-noise limited, I do think we can expect some improvement in the near future. The Sony Alpha 7S changes the conversion gain as you go from ISO 1600 to 2000, dramatically lowering the read noise. Designs that offer more granular adjustments to conversion gain might do what he wishes for.

Jim

[Telecaster]

Thom refers to "pragmatic dynamic range," which I guess takes into account his preferred software and workflow. He also mentions exposure blending and notes the 7Dii & D7100 can yield essentially identical results DR-wise via this method. IOW, in scenes exceeding the single-exposure DR of both cameras it doesn't matter which one you use.

-Dave-

[ErikKaffehr]

Hi Jim,

I would suggest that there is a benefit of having low readout noise, and that would be that blacks and near blacks have little impulse type noise. Those areas, Zone 0 in Ansel Adams speak I guess, don't hold detail but would be ugly with colourful salt & pepper noise. It's a mix of red, green and (have you seen?) blue peppers.

Best regards
Erik

Ps. I need to check your blog again, your writing there is even better than your postings here! :-)

If your metric of dynamic range is the ratio of full-scale to the average signal level that produces a signal-to-noise ratio of around 10 or so, which happens to be mine, without a radical change to the design of the sensor, there's not much progress to be made here. If he's talking about engineering dynamic range, which is read-noise limited, I do think we can expect some improvement in the near future. The Sony Alpha 7S changes the conversion gain as you go from ISO 1600 to 2000, dramatically lowering the read noise. Designs that offer more granular adjustments to conversion gain might do what he wishes for.

Jim

[Jim Kasson]

I would suggest that there is a benefit of having low readout noise, and that would be that blacks and near blacks have little impulse type noise. Those areas, Zone 0 in Ansel Adams speak I guess, don't hold detail but would be ugly with colourful salt & pepper noise. It's a mix of red, green and (have you seen?) blue peppers.

Erik,

You raise a good point. However, if black is what you're looking for, black is pretty easy to get in post. If you want to automate it, you could apply a median filter through a luminance mask. I'm in the middle of a study of the spatial frequency characteristics of read noise. Read noise has in the past had reputation of being ugly, that ugliness stemming from relatively high low-frequency energy compared to the white noise that you get from the photon statistics. One of the things that I'm finding out, at least in the Nikon D810, which is the only camera for which I've fully crunched the data at this point, is that almost all of the low-frequency energy does not change from frame to frame. Computing an average read noise frame and subtracting that from a series of images results in almost all the read noise ugliness disappearing.

More on that as I get more results. Thanks for the compliments about my blog. I do put a little more effort in making the writing in my blog posts stylish. Having no (self-imposed) space limitations also helps.

Thanks,

Jim

[Fine_Art]

You may want to try astronomy software that is capable of using bias frames.

http://en.wikipedia.org/wiki/Bias_frame

[Jim Kasson]

You may want to try astronomy software that is capable of using bias frames.

http://en.wikipedia.org/wiki/Bias_frame

I use Matlab.

[ErikKaffehr]

Jim,

I may agree on that. Something I was thinking about is that raw developers should have a slider for near dar noise reduction.

One of my points is that I am very much in favour of a parametric workflow, where we are working from a raw image and just keep record of the necessary adjustments. Going out to Photoshop increases file size six times and throws away information.

That said, a parametric work flow needs a persistent image format, I don't feel today's raw formats fulfil that requirements, with DNG being a possible exception.

Best regards
Erik

Erik,

You raise a good point. However, if black is what you're looking for, black is pretty easy to get in post. If you want to automate it, you could apply a median filter through a luminance mask. I'm in the middle of a study of the spatial frequency characteristics of read noise. Read noise has in the past had reputation of being ugly, that ugliness stemming from relatively high low-frequency energy compared to the white noise that you get from the photon statistics. One of the things that I'm finding out, at least in the Nikon D810, which is the only camera for which I've fully crunched the data at this point, is that almost all of the low-frequency energy does not change from frame to frame. Computing an average read noise frame and subtracting that from a series of images results in almost all the read noise ugliness disappearing.

More on that as I get more results. Thanks for the compliments about my blog. I do put a little more effort in making the writing in my blog posts stylish. Having no (self-imposed) space limitations also helps.

Thanks,

Jim

[ErikKaffehr]

Hi Jim,

I would say a good tool for engineers, who actually know what they are doing.

I am not really there…

Best regards
Erik

I use Matlab.

[Jim Kasson]

I would say a good tool for engineers, who actually know what they are doing.

I am not really there…

Erik, don't worry. If something comes of this, I'll come up with a way to tame it for ordinary, but tech-savvy photographers. DCRAW does dark-frame subtraction.

Jim

[Colorado David]

Check to see what BIF guys say on Naturescapes.

Link?

[Torbjörn Tapani]

Erik, don't worry. If something comes of this, I'll come up with a way to tame it for ordinary, but tech-savvy photographers. DCRAW does dark-frame subtraction.

Jim

Cameras do dark frame subtraction with long exposure noise reduction. But you are talking about bias frames right? Astrophotography programs calibrate a stack of raws with bias frames, even the dark frames for a master dark.

[dwswager]

Perhaps what's also obvious is that there's more to a camera than the sensor.

Particularly for an action shooter.

Obviously.  We are basically buying 2 things when purchasing a camera.  The ability to get the images we want and the quality of the data that represents that image.  I'm just happy I was forced to develop the technique to get the image with a 6fps camera with better quality at $850 less.

Tip:  With a 7fps camera and a baseball batter, it is possible to take a sequence and never get the ball in the frame, let alone compressed on the bat.  But with good technique, you can get it with 1 frame!  Hint:  Don't watch the bat, watch for the hands to come through the hitting zone.

[Jim Kasson]

Cameras do dark frame subtraction with long exposure noise reduction. But you are talking about bias frames right? Astrophotography programs calibrate a stack of raws with bias frames, even the dark frames for a master dark.

"...even the dark frames for a master dark." I'm not quite getting what your saying. I'm averaging 256 exposures of a dark field to get rid of the frame-to-frame randomness. Then I'm subtracting the averaged image from subsequent ones.

As I understand what LENR on cameras does, it only subtracts one dark frame from another, reducing the static read error, but actually increasing the amount of frame-random noise, since sqrt(firstFrameRandomNoise^2 + secondFrameRandomNoise^2) = 1.414 * firstFrameRandomNoise,  if firstFrameRandomNoise = - secondFrameRandomNoise and the two are uncorrelated. I'm doing the average subtracting with short exposures. It doesn't work on all cameras.

Jim

[Torbjörn Tapani]

Yes exactly. One dark frame actually increase noise but removes unwanted signal. A program like deep sky stacker will work with stacks of different frames. I'm certainly no expert but a typical scenario would be you have 20 (or more) light frames of your deep sky object, then during the night you also grab some 20 dark frames with the same duration, then you take (next day perhaps) some flat field images to remove vignetting. And you also put a lens cap on, fastest shutter speed and grab a series of bias frames. 20+ or 256 in your case. These are averaged together and are used to calibrate all of the other types of frames, basically to remove read pattern noise. Then you end up with master dark and flat field images that you can reuse with different light frames. More serious astrophotographers will keep a library of master darks for different sensor temperatures.

So are you averaging dark frames? I don't know theory and math but I guess the result would be the same more or less. For shorter exposures it might be practical. Dark + bias frames might be used only because it is impractical to grab so many long exposure darks.