December 10, 2016, 07:48:58 PM

Author Topic: DxOMark Sensor Performance: Nikon vs. Canon  (Read 127267 times)

Aglet

  • 5DSR
  • *******
  • Posts: 1371
Re: DxOMark Sensor Performance: Nikon vs. Canon
« Reply #360 on: October 19, 2012, 10:42:13 PM »
I'm just gonna repeat myself in this thread and say I went out last night to shoot night-city-skylines with a D5100 and D800 and used those lovely clean 100 ISO raw files to produce images that absolutely can not be done with the same single shot technique using a Canon body without the Canon produced image showing FPN that would completely wreck the shot.

Until DxO comes up with a way to publish Fixed Pattern Noise in a meaningfully comparable way this battle's likely to wrage on until Canon develops a better imaging system.

FPN is more the issue than absolute DR, is it not?
Yes, they're inter-related but I'll take a 10 stop DR camera with random noise over an 11 or 12 stop DR camera WITH FPN.

Since I happen to have a 13 stop DR camera WITHOUT FPN, and it's dark outside, I SHOULD go play with it! :)

canon rumors FORUM

Re: DxOMark Sensor Performance: Nikon vs. Canon
« Reply #360 on: October 19, 2012, 10:42:13 PM »

bdunbar79

  • Canon EF 400mm f/2.8L IS II
  • *********
  • Posts: 3111
Re: DxOMark Sensor Performance: Nikon vs. Canon
« Reply #361 on: October 19, 2012, 10:43:26 PM »
I'd rather correct information presented in broken grammar than misleading information wonderfully presented with perfect grammar and spelling.... don't judge a book by its cover.

Uh, NEWS FLASH!  We're way, way beyond the cover.
2 x 1DX
B1G, MAC, GLIAC

LetTheRightLensIn

  • Canon EF 400mm f/2.8L IS II
  • *********
  • Posts: 4757
Re: DxOMark Sensor Performance: Nikon vs. Canon
« Reply #362 on: October 19, 2012, 10:59:27 PM »
I'd rather correct information presented in broken grammar than misleading information wonderfully presented with perfect grammar and spelling.... don't judge a book by its cover.

Uh, NEWS FLASH!  We're way, way beyond the cover.

30 pages may as well be the cover when the book has (by the time it will be all written and finished it seems) 30,000 pages  ;)

rpt

  • Canon EF 300mm f/2.8L IS II
  • ********
  • Posts: 2629
Re: DxOMark Sensor Performance: Nikon vs. Canon
« Reply #363 on: October 19, 2012, 11:07:18 PM »
vem orkar läsa allt detta? vi kan ta det på svenska ,  god natt.
Bahai Bhai Mikael thanda dimaak thi vichaar kar. Asa raghla tar kasa chalel? Aapko Kya ham sub se itnee nafrat kyun hai? Tumi khoob beshee korcho.

 :)

Fyi, there are 4 languages there.

Kshama karen, mai ne galat likha tha. O aame theek korechi.
« Last Edit: October 20, 2012, 12:25:54 AM by rpt »

rpt

  • Canon EF 300mm f/2.8L IS II
  • ********
  • Posts: 2629
Re: DxOMark Sensor Performance: Nikon vs. Canon
« Reply #364 on: October 19, 2012, 11:11:03 PM »
I'd rather correct information presented in broken grammar than misleading information wonderfully presented with perfect grammar and spelling.... don't judge a book by its cover.

Uh, NEWS FLASH!  We're way, way beyond the cover.

30 pages may as well be the cover when the book has (by the time it will be all written and finished it seems) 30,000 pages  ;)

:)

Imagination_landB

  • Rebel SL1
  • ***
  • Posts: 96
Re: DxOMark Sensor Performance: Nikon vs. Canon
« Reply #365 on: October 19, 2012, 11:26:03 PM »
Michael risedal is SOOO over us because he knows another language than english. Sérieusement tu tappe sur les nerfs de tout le monde tu devrais aller jouer avec ton kodak et arrêter de déblaterer , c'est un forum pas une bataille sur qui a raison ou tort. T'es vraiment n'importe quoi. I think everyone here is getting tired of you and your useless arguments.
6D, Gripped 60D. 120-300 2.8 OS, 50 1.8, 8-16, 24-70 VC.

rpt

  • Canon EF 300mm f/2.8L IS II
  • ********
  • Posts: 2629
Re: DxOMark Sensor Performance: Nikon vs. Canon
« Reply #366 on: October 19, 2012, 11:31:34 PM »
Michael risedal is SOOO over us because he knows another language than english. Sérieusement tu tappe sur les nerfs de tout le monde tu devrais aller jouer avec ton kodak et arrêter de déblaterer , c'est un forum pas une bataille sur qui a raison ou tort. T'es vraiment n'importe quoi. I think everyone here is getting tired of you and your useless arguments.
And we're on! Yes, I gave a sample of 4. So we now shift fron DR to LR (language range)  :)

canon rumors FORUM

Re: DxOMark Sensor Performance: Nikon vs. Canon
« Reply #366 on: October 19, 2012, 11:31:34 PM »

Imagination_landB

  • Rebel SL1
  • ***
  • Posts: 96
Re: DxOMark Sensor Performance: Nikon vs. Canon
« Reply #367 on: October 19, 2012, 11:31:43 PM »
Aaaaaaand for the topic, I can't wait to see if they can overcome their canon ''hate'' for once. I think we all know Nikon new cameras are great! but their numbers are a little offroad
6D, Gripped 60D. 120-300 2.8 OS, 50 1.8, 8-16, 24-70 VC.

Imagination_landB

  • Rebel SL1
  • ***
  • Posts: 96
Re: DxOMark Sensor Performance: Nikon vs. Canon
« Reply #368 on: October 19, 2012, 11:33:25 PM »
Ha! looks like it's a game for him so why don't we all play to it  ;D
« Last Edit: October 19, 2012, 11:39:09 PM by Imagination_landB »
6D, Gripped 60D. 120-300 2.8 OS, 50 1.8, 8-16, 24-70 VC.

rpt

  • Canon EF 300mm f/2.8L IS II
  • ********
  • Posts: 2629
Re: DxOMark Sensor Performance: Nikon vs. Canon
« Reply #369 on: October 19, 2012, 11:41:39 PM »

And we're on! Yes, I gave a sample of 4. So we now shift fron DR to LR (language range)  :)
Ha! looks like it's a game for him so why don't we all play to it  ;D
[/quote]
  :)


LetTheRightLensIn

  • Canon EF 400mm f/2.8L IS II
  • *********
  • Posts: 4757
Re: DxOMark Sensor Performance: Nikon vs. Canon
« Reply #370 on: October 19, 2012, 11:43:59 PM »
Michael risedal is SOOO over us because he knows another language than english. Sérieusement tu tappe sur les nerfs de tout le monde tu devrais aller jouer avec ton kodak et arrêter de déblaterer , c'est un forum pas une bataille sur qui a raison ou tort. T'es vraiment n'importe quoi. I think everyone here is getting tired of you and your useless arguments.

ne visi

bdunbar79

  • Canon EF 400mm f/2.8L IS II
  • *********
  • Posts: 3111
Re: DxOMark Sensor Performance: Nikon vs. Canon
« Reply #371 on: October 20, 2012, 11:30:30 AM »
I'd rather correct information presented in broken grammar than misleading information wonderfully presented with perfect grammar and spelling.... don't judge a book by its cover.

Uh, NEWS FLASH!  We're way, way beyond the cover.

30 pages may as well be the cover when the book has (by the time it will be all written and finished it seems) 30,000 pages  ;)

A 30,000 page thread would be amazing actually.  That would be impressive!
2 x 1DX
B1G, MAC, GLIAC

TheSuede

  • Canon AE-1
  • ***
  • Posts: 54
Re: DxOMark Sensor Performance: Nikon vs. Canon
« Reply #372 on: October 20, 2012, 12:16:27 PM »
There's quite a lot in here, so I'll break it up and even .../cut/... some parts, and if you feel something has been cut inappropriately, just say so.
I think this is a critical point, and possibly the root of the contention of the "DXO DR naysayers." As someone who prints a lot myself, perhaps I can offer some insight.
Assuming you print at native resolution, printing does not average the original amount of information into something less. .../cut/...
Printing is NEVER a pixel > pixel matter, the screening of the print may be at 2400dpi even though you're printing at 300dpi. Otherwise you would need 16 million differently colored inks to get a full hue/tone presentation. Since you "only" have between four and twelve (we have 10-pigment printers, with two neutral densities) possible inks - and in some head models maybe two different ink spot sizes - you need a variation of [16x10^6 / 8] over area, dithered to get a full tone resolution. I set the divisor to eight, as an average of the amount of available individual inks in a modern printer here. How the screening process (dither or a complete image RIP) is done determines how good the printer handles detail per mm or in the equivalent measurement set: MTF per dpi.
There are other problems with DXO calling their rated DR "Print DR", though. Assuming you are using a godly form of paper, such as Innova FibaPrint Gloss, which has a dMax of over 2.7, you might be able to get 7 stops or so from a print. Your average fine art print paper has a dMax randing from around 1.3 to 1.5 on average to 1.75 or so for some of the more recent higher-end fine art papers. That gets you maybe 5-6 stops of DR.
Don't make the mistake of mixing up tone resolution with DR. They are never the same in any practical application. Example:
A certain measurement has the DR of 10:1 (say a measurement range of 1 to 10), and a resolution of "1" in that range. Ten discrete steps can be clearly differentiated in the original.
A certain presentation type has the linear output range of 10-20. That gives a DR of only "2" since 20 is only two times as much as 10. But the presentation still has ten discrete levels very clearly distinguishable from each other, meaning that the tone resolution hasn't changed. On a visual inspection, you haven't limited the measurement DR, just shifted the base point (and lowered the detail MTF of course).
So that takes us back to the definition of DR. I'm happy to accept that DXO has a purely mathematical interpretation of DR, the ratio between white point (maximum saturation) and black point (noise floor). Again, though, I am not sure it is a useful or realistic definition of what dynamic range is. When one thinks about the value of dynamic range in digital photography, the first thing that usually comes to mind is the ability to recover useful detail from deep shadows. I say from the shadows, as I think any photographer who uses digital knows that it is critical to preserve the highlights, as once they are clipped, detail is well and truly gone.
Realistic; who knows, at least I can tell a whole lot of things about a camera and the resulting images, and what you can DO with the camera - just by knowing the DR and some other base performance figures. You would get the same answer from any other competent machine vision specialist or optoelectrician.
Practical; Since a camera sensor signal is linear you can move around as you want in it, internal contrast will be constant. This means that I can expose (photometric exposure) maybe one full stop less with a camera with good DR, giving me more "practically usable latitude" in both highlight and shadow. This is not a very difficult PP operation - I put (in the raw converter) the exposure compensation at +0.5 and make the highlight tone curve a little less harsh in the cutoff knee.
If the DR is good, I can shorten my shutter speeds or get more DoF (stop down) at low ISOs - without loosing any image quality compared to a low-DR camera used at longer shutter speed or shorter DoF!
More DoF or/and shorter shutter speeds are in most situations something very practical, wouldn't you say?
The dispute on record here, if I may define it according to my own views as well as that which I've read from other DXO DR naysayers, is this:

What value does DXO PrintDR (the mathematically derived ratio between white point (maximum saturation, FWC) and black point (electronic noise floor)) have in a real-world context?

From the standpoint of simply moving the black point in a downsampled image, the only thing that occurs is shadows become darker. One LOSES information during the process of downsampling, so the primary benefit of having additional DR in the hardware no longer applies. In the context of viewing images on a computer screen, primarily done via the web, having a deeper black point might be valuable. Computer screens generally support a much deeper black point than actual prints on paper (particularly prints on high quality fine art paper), although none actually support 14 stops of DR regardless, and the average consumer screen is only 6-bit, so roughly the same DR as a print.
Firstly - Most cheap laptop (and cheap TN) screens use 6-bit with 240Hz time scale (delta/sigma) dither to get 8 bits of tone resolution. None of my devices (except maybe my phone) use lower than 8 discrete bits, and both my TV and my computer screens are true 8-bit >> 10-bit time-dithered.
Secondly - This tone resolution is quantized in a gamma-corrected space, usually around gamma>>2.0. If you look at the sRGB gamma the step between the first 14 (of 256) bits is 1/13 of the bit value. This means the linear DR of 8-bit sRGB in the ideal application is 13x255 = 1:3315 or about 11.5 linear bits/Ev. A well calibrated HD-TV will follow ITU-R BT709, and present a step of 4.5 in the lower part of the gamma-curve - giving a linear DR of 4.5 x (235-16) = 1:985 - or slightly less than 10 bits/Ev
8-bit sRGB as a format standard has almost the same DR per pixel as a 1Dx. (!) -But in a nonlinear tone mapping - that's the difference.

When it comes to real print, assuming one is printing at native size, or an upsampled image, original detail is preserved or slightly softened, but none of it is lost due to downsampling. Regardless, assuming one even does significantly downsample a D800 image so they can print at 8x10", even printed on the highest dMax papers on the market with the brightest L* rating, your going to get HALF the DR you should supposedly be getting from DXO's 14.4 stop Print DR rating.
.../LONG cut/...
At best, DXO's downsampled DR rating should probably be called Web DR. It is not detail-preserving Photographic DR, as upon downsampling you lose detail.  It is definitely not Print DR, since a print is inherently more about color richness and gamut than white-to-black point dynamic range. The depth of blacks sometimes matters in a print, however the deeper your black point in print, the harder it tends to be to actually discern fine shadow detail.
.../cut/...
Again, you're comparing DR and tone resolution as if they were the same thing. And the comparison definitely does not seem coherent in how noise in different tone levels (brightness zones?) is perceived in a real print.

Photography is - when broken down to practical discrete steps - a series of [input DR + tone resolution] to [output DR + tone resolution] translations. As long as the tone resolution of the combined DR+TR of the receiving end is larger than the sending end the transformation can be lossless.
1) 3D object space (reality... :-) ) is projected on to a focal plane (sensor or film) through a lens, where you lose DR due to diffraction, haze and flare. Tone resolution is still infinite, limited only by quantum light physics.
2) the image space (the projection) is to be translated into electrical and then digital signals (the sensor and sensor electronics). DR is lost due to noise issues, TR is lost due to noise and quantization issues.
3) the linear readout has to be tone-mapped into a standardized color space, often a gamma corrected 8-bit RGB space. Here the tonemapping and tone curves determine how much DR you lose - if you lose any DR at all. In a camera phone or a cheap compact, sRGB actually has a much greater DR than what the input can provide! Tone resolution is (often) limited to 1:255 (8bits)
4) the standardized image format has to be rasterized to make it printable. DR is limited by the paper white and ink black densities, tone resolution is limited only by the rasterization scheme.

So, it's quite easy to make a 12bit DR deep shadow detail show up as an easily recognizable noise-free detail even in a 7bit (Ev) presentation DR.

So, what is the value of DXO Print DR? Realistically, practically, physcally...what do I actually gain by downsampling my full-detail RAW into a smaller-sized TIFF? For that matter, what value does DXO Print DR have if I save as a compressed JPEG for viewing on the web? Are we really just talking about a DXO weighted score, and nothing more? If so, should it really be called Dynamic Range, or is there a better term DXO could use that wouldn't come off as some kind of sketchy maneuvering (real or simply perceived) of their results in favor of a major monetary contributor?
Trying to redefine a metric that has been used practically, and for very many practical reasons in very many practical circumstances - by thousands upon thousands of signal processing engineers, sensor developers, imaging software developers (including the guys over at Canon DPP development center) and imaging process logic circuit developers is NOT in any way productive, I'd say it's very counterproductive. Especially since the connection between the measurement value and images in reality is so easy to show.

What would help a lot for most people is to understand what DR is, when used and put in the context that it is MEANT to be used. It plays ONE very important part in the most basic breakdown of the individual parameters that is universally used to measure or determine a signal quality - and a signal quality is the base for image quality assessment.

The camera total raw DR does in very large and noticeable amount have an effect on how the complete chain from object space (reality) to print can be realized. A camera with good DR has the (optional!) ability to show a lot more shadow detail (without adding noise!) in the final result, even if the paper/ink combination is pretty poor.
« Last Edit: October 20, 2012, 12:17:59 PM by TheSuede »

canon rumors FORUM

Re: DxOMark Sensor Performance: Nikon vs. Canon
« Reply #372 on: October 20, 2012, 12:16:27 PM »

LetTheRightLensIn

  • Canon EF 400mm f/2.8L IS II
  • *********
  • Posts: 4757
Re: DxOMark Sensor Performance: Nikon vs. Canon
« Reply #373 on: October 20, 2012, 03:12:22 PM »
Is that the assertion? Or is the assertion that a 36 MP sensor with 13.2 bits of DR at 36 MP should be described as having 36 MP of resolution and 13.2 bits of DR?

Camera 1 has 36 megapixels and 11 stops of dynamic range per pixel, camera 2 has 20megapixels and 12 stops of dynamic range per pixel.

Which camera has more dynamic range if I display or print their images at the same size ? If I downsample the 36mpx image to 20mpx, I will get more than 11 stops of dynamic range, but do i get more than 12 ?

I think this is a pretty valid question, provided you intend to view the whole image on print or screen, as opposed to just viewing 100% crops.

This is a complex question, since what DxO basically fails to include in their DR vs resolution compensation (the "print" view option) is that no Bayer-based image ever contains equal noise energy all the way up to 1/f. The interpolation stage often called the "de-mosaic" stage necessary (remember, two out of the three channels in each pixel has had to be estimated after the capture...) is effectively filtering the higher noise frequencies out, and tends towards zero at 1/f.

In a layman's terms, you could describe this high-frequency filter as:
-"The noise - or average pixel difference - is stronger in power when you compare two pixels a few pixels apart from each other than if you compare two pixels next to each other."

The end result of this is that the first ~30% of downsampling - down to 70% original scale that is - does not lower image noise power [over the image width] by any significant degree. There wasn't much noise energy in the frequency band that we've filtered away, so what we've basically done is to condense the image information.

But then there's also a more subtle effect. The human eye does not react as strongly to fine-grained noise as it does to coarse-grained noise. This does mean that when you MEASURE the average pixel noise power, it might not have been lowered by any appreciable amount - but when you LOOK at the downsampled image you will perceive the image as less noisy anyway.

So - downsampling 36MP to 20MP would theoretically give you an added:
sqrt(36/20) = 1.34 linear scale
>> convert to log2 >>
log(1.34)/log(2) = 0.42Ev or "bits" of DR

But you wouldn't get 0.42Ev in a real, converted image - you would get maybe 0.1Ev and a much tighter (less objectionable) noise pattern.

From this point on downwards however the noise spectrum could be said to be sufficiently close to a linear 1/f behavior, and you would get the full noise-power lowering effect in practice too. So, continuing down in size would yield the full theoretical gains of log(sqrt(linear scale))/log(2). Together with a lowered resolution per image frame of course... :-)

OTOH, if one was trying to compare things for real one might use an advanced, adaptive NR algorithm and not a quick downsample so that might counterbalance some, or maybe even actually more than all, of the loss you mention where the first 30% of down sampling doesn't help to so much due to the debayer already having done some work, etc. So in the end, perhaps just imaging the full amount would apply would give a relatively realistic estimate, not quite sure how it all balances out.

Fishnose

  • Rebel T6i
  • ****
  • Posts: 104
Re: DxOMark Sensor Performance: Nikon vs. Canon
« Reply #374 on: October 20, 2012, 03:59:43 PM »
Suede, you nailed it. To the wall.

canon rumors FORUM

Re: DxOMark Sensor Performance: Nikon vs. Canon
« Reply #374 on: October 20, 2012, 03:59:43 PM »