Pages: 1 [2] 3   Go Down

Author Topic: DXO Mark - 5DS / R - Best Canon Sensor Yet.  (Read 24859 times)

dwswager

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1375
Re: DXO Mark - 5DS / R - Best Canon Sensor Yet.
« Reply #20 on: July 10, 2015, 10:58:19 pm »

Hi,

No, that's not correct. The shadows (and thus the DR) of both cameras are virtually identical at nominal ISO 800 or higher (see attachment, normalized to same size output). When shot at ISO 100 and pushed 8x in post, yes then the Nikon is cleaner, but that would be a silly thing to do with any Canon. Once at ISO 800-1250, one can safely push in post processing, with little different  versus shooting at higher  ISOs, other than gaining more highlight headroom when the ISO is not cranked up too much.

Cheers,
Bart

App;es and Oranges.  Try shooting a Canon with 3 stops underexposure and then process.  Hence, you must shoot the Canon at 2-3 stops more ISO than the Nikon.
Logged

Bart_van_der_Wolf

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 8901
Re: DXO Mark - 5DS / R - Best Canon Sensor Yet.
« Reply #21 on: July 11, 2015, 03:53:27 am »

App;es and Oranges.  Try shooting a Canon with 3 stops underexposure and then process.

Hi,

Maybe I didn't explain it well enough, but shooting at ISO 800, is shooting with 3 stops under-exposure. I see no reason to shoot at 6 stops underexposure if -3EV will do it better. And from the looks of it, using anything over ISO 800-1250 gains little quality, except that it produces a brighter embedded thumbnail.

Virtually all that boosting the ISO does in a mostly ISO invariant camera like the Nikon 810 is adjust the camera's exposure meter towards underexposure. The camera's firmware settings and the Raw converter then adjust for that underexposure during conversion. The same happens in a Canon, but the Canon sensor is ISO variant because it will additionally adjust the ADC amplification (gain) to reduce the amount of read-noise, which as we know is needed due to the relatively poor read-noise performance at low ISOs compared to modern Sony sensors.

Cheers,
Bart
Logged
== If you do what you did, you'll get what you got. ==

Bart_van_der_Wolf

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 8901
Re: DXO Mark - 5DS / R - Best Canon Sensor Yet.
« Reply #22 on: July 11, 2015, 04:26:26 am »

Perhaps people don't talk about colour so much as it can be dealt with in raw development. Role your own profiles to meet your own requirement etc

Hi Dave,

That's only partly true. Metamerism is something that is caused by the specific spectral transmission characteristics of the Bayer CFA filter layers and the spectral properties of the color it samples. This will cause two subjects with the same colors to be interpreted differently depending on the illumination and reflection spectra when shooting that subject. A color e.g that has spectrally pure yellow color, and a color that is a mix of red and green light, may look to a human observer like they have the same color, but the camera may see a difference (a 'wrong' color representation).

Skin color is e.g. rather difficult, because it reflects a mix of reds and yellows and Infra-red that locally varies as relatively more or less blood is directed more to the surface or more to deeper tissues, and pigmentation and skin thickness add more variation.

A profile as such will be able to reduce (but not totally eliminate) all of the errors, but usually only for a specific illumination.

The Bayer CFA filters are also not spectrally perfect filters, they have secondary absorptions and transmissions, which will contaminate colors with different wavelengths than the R/G/B pass band that is supposed to be sampled. For example Blue light will register as a bit of Red, and Infrared light will register a muddy brown because it also adds blue and green signal, thus desaturating and changing the other colors. This will not be separable after the pollution has taken place. Just like you cannot unscramble an omelette.

Cheers,
Bart
Logged
== If you do what you did, you'll get what you got. ==

ErikKaffehr

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 11310
    • Echophoto
Re: DXO Mark - 5DS / R - Best Canon Sensor Yet.
« Reply #23 on: July 11, 2015, 04:38:51 am »

Hi,

The SMI essentially measures how well a set of ColorChecker patches can be reproduced by an optimal "compromise matrix" for a given illuminant. In a sense it is a bit similar to say how far the sensor is from Luther Ives condition.

I don't know how relevant SMI is for colour rendition, though. I had some discussion on the issue with Tim Parkin, editor of OnLandscape, he clearly finds a very good correlation between colour reproduction and SMI. Iliah Borg who is a well respected developer of raw processing tools (RawDigger, LibRaw etc) doesn't feel SMI is very relevant. Anders Torger, who has done a lot of good development on color profile generation recently, also seems to feel that SMI is perhaps not a very good measure.

In addition, it may be that CFA plays an important role separation or lack thereof of certain shifts green as almost all CFA-s have very steep gradients between 550 and 600 nm. So it may be that a certain CFA is benefitial to reproduction of skin and another one may be better at reproducing say chlorophyll.

Ultimately, colour profiles probably dominate colour reproduction.

I would say that the significant improvement of SMI pared with the limited high ISO settings on the 5Ds may indicate colour reproduction was at higher priority than good high ISO performance.

Best regards
Erik

Hi,

Do not misinterpret the numbers, and yes they need interpretation.

Color sensitivity as measured by DxOmark, is very much signal to noise ratio and dynamic range driven. And we know that Canon sensors are not leading in that aspect. Also, the small photosites of the 5DS with their smaller 'well depth' or storage capacity will have a harder time to collect many photons. But they compensate for their small size by their large number of photosites per unit area.

Now look at the Sensitivity metamerism index (ISO 17321): 5d Mark III was 74 / 75 (for CIE-D50 / CIE-A), the 5DS is 82 / 71, and e.g. the D810 is 80 / 78. So the Daylight metamerism index has improved significantly over its predecessor.

Cheers,
Bart
Logged
Erik Kaffehr
 

Bart_van_der_Wolf

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 8901
Re: DXO Mark - 5DS / R - Best Canon Sensor Yet.
« Reply #24 on: July 11, 2015, 04:59:34 am »

It's really strange, though. The 5DsR has the exact same metamerism results as the 5D3: 74/75 (CIE-50/CIE-A). But the 5Ds is much better for daylight at 82/71, which is close to the improvement I was hoping for.

Hi Jeff,

Good catch, I hadn't compared the scores between the two models yet. This is indeed a strange difference, much larger than one would expect from the Color Response data from the two models, which are just slightly different but overall quite similar.

Quote
How can this be? Did Canon tweak the 5DsR for artificial light and the 5Ds for daylight, without telling us?

Or did DxO have a bad copy of one of the cameras, or made an error?

I don't know. Time will tell if it is corrected/adjusted, or not.

Quote
There are three theories I've been considering.

The first is that Canon has weakened the CFAs on their latest cameras to let more light in to get lower noise at high ISOs, compromising accurate color to do so. But Canon doesn't let on what they are doing. The metamerism scores seem to be the best indicator. The 5Ds/5DsR was rumored to have stronger CFAs.

Yes, over the years there was a trend of using less selective CFAs, more spectral pass-band overlap and more transparent, to increase the total amount of light reaching thee sensor. That higher transmission would allow higher sensitivity, and thus less need to push the ISO because that would boost noise. An overall boost of Quantum efficiency would improve the noise statistics, but at a cost of less accurate color separation.

This was supposed to be improved a bit in the 5DS series, towards better color separation. The 5DS metamerism sensitivity indicates that improvement, but the 5DS R not. Strange.

Quote
Nikon used the same AA filter cancellation method in the D800/D800E, but DxO's metamerism scores were only 1 point different. So the AA filter can't be causing this huge metamerism score difference in these new 5Ds cameras.

Canon has been totally silent about what they've been doing to color, so I wouldn't expect them to announce color changes to these cameras (though, it would be very helpful). But why would the AA version have better daylight results than the cancelled AA version?

I would also not expect the AA-filtering to play a role, unless the different orientation (or replacement) of one of the birefringent OLPF layers also changed the spectral transmission characteristics of the filter stack. Anyway, it's more likely that the filter stack has something to do with it, assuming it is not an erroneous number in their report, than that they would have to run two separate productions of CFA filtered sensors.

Cheers,
Bart
« Last Edit: July 11, 2015, 05:33:03 am by BartvanderWolf »
Logged
== If you do what you did, you'll get what you got. ==

Bart_van_der_Wolf

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 8901
Re: DXO Mark - 5DS / R - Best Canon Sensor Yet.
« Reply #25 on: July 11, 2015, 05:43:27 am »

Ultimately, colour profiles probably dominate colour reproduction.

I would say that the significant improvement of SMI pared with the limited high ISO settings on the 5Ds may indicate colour reproduction was at higher priority than good high ISO performance.

Yes, color profiles remains important, but they have to build on what was already captured as an abstraction of reality. Given that that tri-chromatic abstraction of a continuous spectrum is also different from human vision (the mismatch with the Luther Ives condition), we can only hope to reduce the overall errors (or tweak for a more specific/limited range of colors), but it won't be perfect. The design goal trade-offs for the 5DS series were apparently geared towards better color separation, not higher ISO, in addition to higher resolution and better MTF from existing lenses.

Cheers,
Bart
Logged
== If you do what you did, you'll get what you got. ==

shadowblade

  • Sr. Member
  • ****
  • Online Online
  • Posts: 2794
Re: DXO Mark - 5DS / R - Best Canon Sensor Yet.
« Reply #26 on: July 11, 2015, 07:43:41 am »

Yes, color profiles remains important, but they have to build on what was already captured as an abstraction of reality. Given that that tri-chromatic abstraction of a continuous spectrum is also different from human vision (the mismatch with the Luther Ives condition), we can only hope to reduce the overall errors (or tweak for a more specific/limited range of colors), but it won't be perfect. The design goal trade-offs for the 5DS series were apparently geared towards better color separation, not higher ISO, in addition to higher resolution and better MTF from existing lenses.

Cheers,
Bart

Human vision is also trichromatic (red, green and blue cones), with significant overlap in the spectral sensitivity of each type. What the human brain does with the sensor information from the eye varies between people, and may or may not be similar to what the camera does.

The only way to get more spectral accuracy would be to sample in many more spectral bands (i.e. instead of trichromatic, to sample light in 10nm, 5nm or even 1nm bands), allowing you to determine, for instance, if the 'green' light you're seeing is actually green, or a mixture of blue and yellow. Barring some sort of quantum mechanical effect where you could sample the entire visible spectrum at the same time and still distinguish the wavelength of each photon (not impossible, but would require a completely different sensor type) this would mean filtering the incoming light and sampling each band sequentially; this would add up to extremely long exposure times.
Logged

ErikKaffehr

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 11310
    • Echophoto
Re: DXO Mark - 5DS / R - Best Canon Sensor Yet.
« Reply #27 on: July 11, 2015, 08:32:01 am »

Hi Bart,

I don't think that overlap is a problem, really. The M and L curves in human vision have a lot of overlap. I would suggest that the steepness of the gradients may matter a lot. I would also suggest IR and UV filtering have a great importance. Chlorophyll, for instance, has a very reflection of IR, a weak IR filter would add a lot of IR to the L-channel.

This page shows a lot of colour renditions of some flowers: http://echophoto.dnsalias.net/ekr/Articles/OLS_OnColor/SimpleCase/index.html , just two images but different profiles. The two last fields are samples from the save flower using spectral data. Enclosed is a spectral plot of one of the green samples.

Best regards
Erik




Yes, over the years there was a trend of using less selective CFAs, more spectral pass-band overlap and more transparent, to increase the total amount of light reaching thee sensor. That higher transmission would allow higher sensitivity, and thus less need to push the ISO because that would boost noise. An overall boost of Quantum efficiency would improve the noise statistics, but at a cost of less accurate color separation.


« Last Edit: July 11, 2015, 08:44:54 am by ErikKaffehr »
Logged
Erik Kaffehr
 

Bart_van_der_Wolf

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 8901
Re: DXO Mark - 5DS / R - Best Canon Sensor Yet.
« Reply #28 on: July 11, 2015, 09:18:28 am »

I don't think that overlap is a problem, really. The M and L curves in human vision have a lot of overlap. I would suggest that the steepness of the gradients may matter a lot.

Hi Erik,

Mathematically it's not difficult, one just need larger negative elements in the color conversion matrix. But subtracting signals will increase noise, and that may interfere with the demosaicing. So a balance needs to be found between selectivity and overlap+noise.

The Human Visual System (HVS) does it chemically (and at different peak wavelengths), which is of course different from discrete number math. It also uses random placement of the receptors, and uses many more densely packed color receptors (cones) near the Fovea (where it matters most) than luminance receptors (rods).

Quote
I would also suggest IR and UV filtering have a great importance. Chlorophyll, for instance, has a very reflection of IR, a weak IR filter would add a lot of IR to the L-channel.

Yes, Chlorophyll contributes IR reflection to all R/G/B filtered photosites because the filter dyes are transparent to IR, so the efficiency of the camera's IR filter in the filterstack makes a difference. Even the Blue color plane scores a signal caused by IR. Human skin also (diffusely) reflects IR, or it absorbs visible wavelengths more.  UV is fortunately largely absorbed by our lenses, and the sensor is also less/not sensitive to wavelengths below 380/350 nm.

Quote
This page shows a lot of colour renditions of some flowers: http://echophoto.dnsalias.net/ekr/Articles/OLS_OnColor/SimpleCase/index.html , just twi images but different profiles. The two last fields are samples from the save flower using spectral data.

Many flowers have wild reflection spectra outside the human visible spectrum, because they rely on insects which can see those wavelength differences better. They can also have some extremely saturated colors, with almost zero reflection in some wavelengths and full reflection in other wavelengths. They are a challenge to photograph and render their colors as we humans see them.

Cheers,
Bart
Logged
== If you do what you did, you'll get what you got. ==

ErikKaffehr

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 11310
    • Echophoto
Re: DXO Mark - 5DS / R - Best Canon Sensor Yet.
« Reply #29 on: July 11, 2015, 09:52:15 am »

Hi,

Yes, they are difficult to reproduce, a very good reason to include them in any evaluation. You know: "we don't do this because it is easy, we do it because it is hard…"-

Best regards
Erik

Hi Erik,

Many flowers have wild reflection spectra outside the human visible spectrum, because they rely on insects which can see those wavelength differences better. They can also have some extremely saturated colors, with almost zero reflection in some wavelengths and full reflection in other wavelengths. They are a challenge to photograph and render their colors as we humans see them.

Cheers,
Bart
Logged
Erik Kaffehr
 

Roscolo

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 728
Re: DXO Mark - 5DS / R - Best Canon Sensor Yet.
« Reply #30 on: July 11, 2015, 04:56:25 pm »

Those DxO test results just made waiting to see the new offering from Sony a VERY easy decision for me. Apparently the Sony will be several hundred dollars cheaper as well. Canon may need to lower the price significantly on the DS R to get me off the fence.
Logged

jeffreybehr

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 339
    • http://community.webshots.com/user/jeffreybehr
Re: DXO Mark - 5DS / R - Best Canon Sensor Yet.
« Reply #31 on: July 11, 2015, 05:00:42 pm »

I'm sure like most Canon cameras it is totally adequate in performance for most people. But it does reinforce the argument that has been growing  about the the Canon 5 range that Canon is lagging in technical / IP terms for the last few years. Whether this is lack of interest, lack of ability or simply a totally dominant sales department who knows ?

But it does give a new take on less is more.

Actually, more is more.
Logged
Obsessed landscape fotografer
http://

Jack Hogan

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 797
    • Hikes -more than strolls- with my dog
Re: DXO Mark - 5DS / R - Best Canon Sensor Yet.
« Reply #32 on: July 11, 2015, 05:27:07 pm »

Hi,

The SMI essentially measures how well a set of ColorChecker patches can be reproduced by an optimal "compromise matrix" for a given illuminant. In a sense it is a bit similar to say how far the sensor is from Luther Ives condition.

I don't know how relevant SMI is for colour rendition, though. I had some discussion on the issue with Tim Parkin, editor of OnLandscape, he clearly finds a very good correlation between colour reproduction and SMI. Iliah Borg who is a well respected developer of raw processing tools (RawDigger, LibRaw etc) doesn't feel SMI is very relevant. Anders Torger, who has done a lot of good development on color profile generation recently, also seems to feel that SMI is perhaps not a very good measure.

In addition, it may be that CFA plays an important role separation or lack thereof of certain shifts green as almost all CFA-s have very steep gradients between 550 and 600 nm. So it may be that a certain CFA is benefitial to reproduction of skin and another one may be better at reproducing say chlorophyll.

Ultimately, colour profiles probably dominate colour reproduction.

I would say that the significant improvement of SMI pared with the limited high ISO settings on the 5Ds may indicate colour reproduction was at higher priority than good high ISO performance.

Interesting take, Erik.  For reference, the difference in effective QE tends to be around 20% +/- 2% or about 1/4 of a stop from one end of the range to the other at the same exposure.  This is usually, but not always, reflected in the camera's Sensitivity (Measured ISO in DxO speak).

Jack
Logged

capital

  • Full Member
  • ***
  • Offline Offline
  • Posts: 222
    • Website
Re: DXO Mark - 5DS / R - Best Canon Sensor Yet.
« Reply #33 on: July 11, 2015, 06:00:44 pm »

Hi,

The colour sensivity is more related to noise than to CFA. There is another figure called SMI that measures colour reproduction potential.

But, colour reproduction may be much more dependent on camera profiles than on CFA design.

Best regards
Erik


Hello!
Based upon DXO, which cites, they equate a correlation of color sensitivity and color depth. Certainly, Canon was stating that the new CFA would help in this regard, but from this particular metric no real change is seen.  “a much stronger CFA [color filter array] which will produce much greater color accuracy than anything currently in the Canon lineup.”
Logged

dwswager

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1375
Re: DXO Mark - 5DS / R - Best Canon Sensor Yet.
« Reply #34 on: July 12, 2015, 10:14:01 am »

Hi,

Maybe I didn't explain it well enough, but shooting at ISO 800, is shooting with 3 stops under-exposure. I see no reason to shoot at 6 stops underexposure if -3EV will do it better. And from the looks of it, using anything over ISO 800-1250 gains little quality, except that it produces a brighter embedded thumbnail.

Virtually all that boosting the ISO does in a mostly ISO invariant camera like the Nikon 810 is adjust the camera's exposure meter towards underexposure. The camera's firmware settings and the Raw converter then adjust for that underexposure during conversion. The same happens in a Canon, but the Canon sensor is ISO variant because it will additionally adjust the ADC amplification (gain) to reduce the amount of read-noise, which as we know is needed due to the relatively poor read-noise performance at low ISOs compared to modern Sony sensors.

Cheers,
Bart

Not sure if we are talking past each other or what.  What I am saying is that you don't dare shoot a Canon (except maybe the 1Dx) at -3 or -4 EV because of the shadows while the Sony sensor makes this a valid option.  Hence, in a situation when you don't have enough light to shoot at the shutter/aperture you need, you can compensate on the Nikon by dialing in under exposure.  As long as  you don't mind the dark thumbnail, you can go quite low.  The only Canon cameras I see at night on local sports fields are the 1Dx and the 6D.  During the day, I see all kinds of Canon cameras, but mainly the 7D and 7DmkII and 1Dx.  I have acquaintances that are quite happy to sit out the night games as they say it just isn't worth the effort.  Of course, the professionals that shoot Canon have the proper equipment to do so at night...the 1DX.  Of course, there is always a penalty you pay, but it is an effective way to get reasonable quality.
Logged

Bart_van_der_Wolf

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 8901
Re: DXO Mark - 5DS / R - Best Canon Sensor Yet.
« Reply #35 on: July 12, 2015, 12:32:51 pm »

Not sure if we are talking past each other or what.  What I am saying is that you don't dare shoot a Canon (except maybe the 1Dx) at -3 or -4 EV because of the shadows while the Sony sensor makes this a valid option.  Hence, in a situation when you don't have enough light to shoot at the shutter/aperture you need, you can compensate on the Nikon by dialing in under exposure.  As long as  you don't mind the dark thumbnail, you can go quite low.

That's correct, thanks to the very clean shadows you can still push a lot in postprocessing.

The strategy on the Canon, although it's not a sports or high ISO optimized camera, is that it's better to push the ISO (at least to ISO 1250) instead of underexposing. That will produce the relatively cleanest shadows that the Canon is capable of, but there will be more noise than with the Nikon if you push that result in postprocessing. But that's not news. The Canon does have a lot of pixels, so downsampling will reduce the noise somewhat.

Quote
The only Canon cameras I see at night on local sports fields are the 1Dx and the 6D.  During the day, I see all kinds of Canon cameras, but mainly the 7D and 7DmkII and 1Dx.  I have acquaintances that are quite happy to sit out the night games as they say it just isn't worth the effort.  Of course, the professionals that shoot Canon have the proper equipment to do so at night...the 1DX.  Of course, there is always a penalty you pay, but it is an effective way to get reasonable quality.

Yes, horses for courses (but the 5DS is not really intended for sports, although it might come a long way with downsampling). And having lenses that (have predictive/tracking) focus well is of course also important in sports and with longer focal lengths.

Cheers,
Bart
« Last Edit: July 12, 2015, 12:34:37 pm by BartvanderWolf »
Logged
== If you do what you did, you'll get what you got. ==

Hans Kruse

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2102
    • Hans Kruse Photography
Re: DXO Mark - 5DS / R - Best Canon Sensor Yet.
« Reply #36 on: July 12, 2015, 02:01:00 pm »

The Canon versus Sony sensors debate is getting a bit long in the tooth, right :)

With the really good HDR function in Lightroom this is become an almost non issue. However there is no excuse for the lower than average performance of the Canon sensor. Shooting both Canon and Nikon I prefer the Canon body design and the Canon lenses, so if Canon suddenly had winning sensors there would be no contest. But I would still shoot both.

Bart_van_der_Wolf

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 8901
Re: DXO Mark - 5DS / R - Best Canon Sensor Yet.
« Reply #37 on: July 12, 2015, 02:13:50 pm »

The Canon versus Sony sensors debate is getting a bit long in the tooth, right :)

Right! There's more to a camera than the DR, and exposure bracketing (if practical, so no sports) will beat any underexposed image in noise quality. Photons rule! Relatively stationary subjects are good candidates for that, and if done with skill, it's only the highlights that need to be masked in, as you have shown in other threads, so subject motion in landscapes is not always a problem. And then there is also great noise reduction software, like Topaz Denoise or DxOs PRIME method.

And then there will be the 'Magic Lantern' magic, for potentially even more quality right out of the camera.

Cheers,
Bart
Logged
== If you do what you did, you'll get what you got. ==

peterottaway

  • Jr. Member
  • **
  • Offline Offline
  • Posts: 91
Re: DXO Mark - 5DS / R - Best Canon Sensor Yet.
« Reply #38 on: July 12, 2015, 09:40:15 pm »

My question is now as over the last few years - why should Canon buyers be expected to pay good money out of their pockets for an inferior product ?

Yes there are plenty of dodges and software costing extra time and money but they all are compromises that affect potential image quality. Whether this bothers you is up to you to decide.
Logged

Jeff Fenske

  • Newbie
  • *
  • Offline Offline
  • Posts: 7
Re: DXO Mark - 5DS / R - Best Canon Sensor Yet.
« Reply #39 on: July 13, 2015, 03:57:26 am »

Hi Dave,

That's only partly true. Metamerism is something that is caused by the specific spectral transmission characteristics of the Bayer CFA filter layers and the spectral properties of the color it samples. This will cause two subjects with the same colors to be interpreted differently depending on the illumination and reflection spectra when shooting that subject. A color e.g that has spectrally pure yellow color, and a color that is a mix of red and green light, may look to a human observer like they have the same color, but the camera may see a difference (a 'wrong' color representation).

Skin color is e.g. rather difficult, because it reflects a mix of reds and yellows and Infra-red that locally varies as relatively more or less blood is directed more to the surface or more to deeper tissues, and pigmentation and skin thickness add more variation.

A profile as such will be able to reduce (but not totally eliminate) all of the errors, but usually only for a specific illumination.

The Bayer CFA filters are also not spectrally perfect filters, they have secondary absorptions and transmissions, which will contaminate colors with different wavelengths than the R/G/B pass band that is supposed to be sampled. For example Blue light will register as a bit of Red, and Infrared light will register a muddy brown because it also adds blue and green signal, thus desaturating and changing the other colors. This will not be separable after the pollution has taken place. Just like you cannot unscramble an omelette.

Cheers,
Bart

So if a weaker CFA is used to let more light into the camera (so ISO can be increased without increasing noise), from what I understand, colors will be less differentiated and less accurate, and that can't be changed with computer software, because "you cannot unscramble an omelette?"

Is that true?

And if the colors in the CFAs are altered for the camera to perform better in artificial light, we can try to counter that in post, but what's done permanently in the CFAs is probably hard to change?

And the same would be true for how they tweak the CFAs to get "more pleasing" skin tones; even though, this would not help landscape colors?

It's hard to find any hard evidence proving this, but this Luminous Landscape article first alerted me of the weakened color filter array problem, and that helped explain to me what was going on with my 5D3 (74 metamerism score), which had color issues that I didn't notice on my 5D2 (80 metamerism score).

[[[As one example, the selection of a CFA, the color pattern put in front of the sensor, is a choice between quality of color, and ISO performance. If the CFA allows each pixel to see a broader spectrum of color (e.g. for the green pixels to see a bit further into yellow) a camera’s ISO range can be modestly increased. The resulting loss in color quality is subtle – subtle variations in color are missed and a handful of specific colors become difficult to photograph.

In a market where a ISO 25,600 camera has a leg up on a ISO12,800 camera, the engineers are under enormous pressure to pick the modestly increased ISO over subtle color quality.]]]

This article is also interesting, and apparently Sensorjen derives his QE numbers from DxO data. Not sure.

[[[However, it is not merely the total amount of light that falls on the sensor that determines the noise in the photo, but also how efficient the sensor is. The primary attributes of sensor efficiency are the QE (Quantum Efficiency -- the proportion of photons falling on the sensor that are converted into electrons) and the read noise (the additional noise added by the sensor and supporting hardware), although it should be noted that a manufacturer may use a weaker CFA to increase the QE, which will reduce the luminance noise at the expense of increasing chroma noise. For example, two sensors may have the same QE and read noise, but one sensor may have a weaker CFA than the other, making it a more noisy sensor. Sensorgen is an excellent resource for sensor QE and read noise, although it does not give information about the CFAs.

However, while QE and read noise are the primary attributes to sensor efficiency, there is more to the story, such as the Bayer CFA (color filter array) which most all cameras use. The Bayer CFA is a color filter covering the sensor in an RGGB pattern. This means that 25% of the pixels are covered with red filters, 50% are covered with green filters, and 25% are covered with blue filters. Of course, the filters actually accept a range of colors, which overlap (otherwise, yellow photons, for example, would never make it through the color filters). How much the filters overlap, and how strong the filters are, also contribute to the sensor efficiency.

For example, the green filter may only admit 60% of the green light that falls on it, but also admit 10% of other colors that fall on it. If we use a weaker filter to increase the transmissivity, we will reduce the luminance noise (more total light will pass through the filter and onto the sensor), but also increase the transmission error by concomitantly allowing a greater percentage of other colors to also pass through. Thus, different manufacturers may strike different balances between luminance noise vs color noise in their choice of color filters. For example, it has been argued that Canon has been steadily "weakening" their CFAs to increase the QE of their sensors. The 5D has a QE of 25%, the 5D2 has a QE of 33%, and the 5D3 has a QE of 49%, as the metamerism index has steadily declined from 84 to 80 to 74, respectively.]]]

I had always thought that Canon would never mess with low ISO performance shot in natural daylight (CRI of ~100), but it looks like I was very wrong.

Canon isn't saying, and few seem to understand what they've been doing. I don't know of even one article that addresses this topic only.
« Last Edit: July 13, 2015, 04:06:33 am by Jeff Fenske »
Logged
Pages: 1 [2] 3   Go Up