Some spectral plots…

[ErikKaffehr]

Hi,

Anders Torger found a site with spectral data for some sensors.

Let's start with comparing Phase One (unspecfied) with H2D (unspecified) and Sony NEX5N.

Observations:

• P1 has low sensivity to blue and high sensivity to red. Balanced for daylight/strobe
• Red filter curve on P1 has 50% transmission at 560nm, NEX5N at 570 and H2D at 580 nm
• Red filter is steepest on NEX5N
• Green curve has same peak and similar FWHM (full with half maximum) for all tree. But Sony sensor has extended green sensivity down to 410 nm.

[ErikKaffehr]

Now let's look at the Canon 5DIII.

The Canon has also a red channel cut off at around 590 nm, but it has a steeper gradient. The "red channel" sensitivity has a tail going down to 510 mm. My guess is that this would give more subtle colour, more like Kodachrome than Velvia look, I would guess. But alas, I may be wrong.

Green sensitivity peaks  490 nm while on the CCD backs it is more 510 nm.

No firm conclusions, but I would think the Canon CFA design  would have the potential to yield a more natural colour and the MFD backs would produce more saturated colours. (Think Canon -> Kodachrome and MFD -> Velvia).

Hi,

Anders Torger found a site with spectral data for some sensors.

Let's start with comparing Phase One (unspecfied) with H2D (unspecified) and Sony NEX5N.

Observations:

• P1 has low sensivity to blue and high sensivity to red. Balanced for daylight/strobe
• Red filter curve on P1 has 50% transmission at 560nm, NEX5N at 570 and H2D at 580 nm
• Red filter is steepest on NEX5N
• Greem curve has same peak and similar FWHM (full with half maximum) for all tree. But Sony sensor has extended green sensivity down to 410 nm.

[ErikKaffehr]

More is coming…

Just to mention, I have compared spectral plots from the Sony NEX5N and the Nikon D2X. They were very close, so I think Sony has the same recipe for high end pro and customer cameras.

Here is Anders Torger's link to the camera SSFs: http://www.ludd.ltu.se/~torger/dcamprof.html#ssf_links

Best regards
Erik

[AlterEgo]

there is one more = http://www.image-engineering.de/content/library/diploma_thesis/christian_mauer_spectral_response.pdf , there you find one more H SSF/CMF

[AlterEgo]

found a site with spectral data for some sensors.

"found" credit to = http://forum.luminous-landscape.com/index.php?action=profile;u=75607 ( http://forum.luminous-landscape.com/index.php?topic=100015.msg819108#msg819108 )

[ErikKaffehr]

Hi AlterEgo,

Your postings are always an interesting source of information!

Best regards
Erik

"found" credit to = http://forum.luminous-landscape.com/index.php?action=profile;u=75607 ( http://forum.luminous-landscape.com/index.php?topic=100015.msg819108#msg819108 )

[AlterEgo]

Your postings are always an interesting source of information!

I wish... I, myself, run into Trantor's SSF/CMF experiments with profiles from SSF/CMF around early 2014 @~ http://forum.rudtp.ru - but naturally that was RU-language forums... he ventured into some color grading/correction nowadays = http://nuclearlight.net/manual.html

[torger]

No firm conclusions, but I would think the Canon CFA design  would have the potential to yield a more natural colour and the MFD backs would produce more saturated colours. (Think Canon -> Kodachrome and MFD -> Velvia).

MFD in this case is Kodak if I'm not mistaken, and those were saturated and quite subjective color. Less overlapping filters, more saturated colors but also more difficult to make neutral color. More overlapping filters less saturated colors, but easier to make neutral color. More overlapping filters however also means that you need to "separate" the colors more in the camera profile which can exaggerate noise. So there's always a tradeoff, but it's not likely we'll see Kodak type of CFAs again in a high end camera. I'm guessing that mobile phone cameras or other high noise cameras still have less overlapping color filters.

Today most cameras have well-overlapping filters and you can make them look very similar with a profile. For extreme range colors there are differences though, and of what I've seen so far in profile design I think Canon's CFAs are more well-balanced than say Sony's for handling high saturation colors. Sony has some sensitivity in blue that makes saturated blues difficult to reproduce realistically. Blue is sort of a problem color as our eyes are not so sensitive in that range. The standard solution is to reproduce blues much lighter than what would be realistic.

There's also a tradeoff to make concerning if the camera should work well in daylight (blue-heavy) or tungsten (red-heavy) type of light. The Sony blue problem is reduced the lower the light temperature as the "over-sensitive" blue response is then reduced.

It would be interesting to have measurements of more modern MFD cameras, but I haven't found any publicly available sources for that.

[ErikKaffehr]

Hi,

My guess when I saw those curves was that the high blue sensitivity is intended to improve handling of available light that often has lower blue content.

I also noted that Sony sensors have a very wide response in the blue channel, downwards to 410 nm.

Another thing worth noting is that Nikon D2X has a very similar curve to the NEX5N. Sony Sony has very similar CFA-s on both a very old Pro Nikon and a much more recent mirrorless APS-C.

Best regards
Erik

MFD in this case is Kodak if I'm not mistaken, and those were saturated and quite subjective color. Less overlapping filters, more saturated colors but also more difficult to make neutral color. More overlapping filters less saturated colors, but easier to make neutral color. More overlapping filters however also means that you need to "separate" the colors more in the camera profile which can exaggerate noise. So there's always a tradeoff, but it's not likely we'll see Kodak type of CFAs again in a high end camera. I'm guessing that mobile phone cameras or other high noise cameras still have less overlapping color filters.

Today most cameras have well-overlapping filters and you can make them look very similar with a profile. For extreme range colors there are differences though, and of what I've seen so far in profile design I think Canon's CFAs are more well-balanced than say Sony's for handling high saturation colors. Sony has some sensitivity in blue that makes saturated blues difficult to reproduce realistically. Blue is sort of a problem color as our eyes are not so sensitive in that range. The standard solution is to reproduce blues much lighter than what would be realistic.

There's also a tradeoff to make concerning if the camera should work well in daylight (blue-heavy) or tungsten (red-heavy) type of light. The Sony blue problem is reduced the lower the light temperature as the "over-sensitive" blue response is then reduced.

It would be interesting to have measurements of more modern MFD cameras, but I haven't found any publicly available sources for that.

[hjulenissen]

...My guess is that this would give more subtle colour, more like Kodachrome than Velvia look, I would guess. But alas, I may be wrong....

Before or after color correction matrix?

The difficulty that I have comprehending such spectral plots is that:
1) Any color that can be uniquely recorded can be made into any other color (using a 3-d lookup). There are good, practical reasons why this should not be done without limits.
2) Spectral selectivity that is far from the desired output would need "lots" of correction. Multiplying the 3 channels by a "nasty" 3x3 correction matrix would increase the likelihood of noise amplification viewed numerically. But a noise of "n" (as measured) might not map into a perceptual correlate very well.
3) A "bad" CFA colorwise might have other properties (avoid attenuating luminance, stable behaviour over time, ...) that makes it "good".

So it would seem that there is some room for optimizing the CFA for various tasks but I fail to see how the spectral selectivity can be analyzed directly without some accompanying color correction matrix/function in order to see the true trade-offs. At least for non-color-scientists.

-h

[ErikKaffehr]

Hi,

Spectral plots are recorded using a monochromator. So you expose frame after frame using near monochromatic light. A monochromator is an affordable device, but it needs to be calibrated by simultaneous reading using a photospectrometer and those are expensive.

The value of this plots it that they show the characteristics of the CFA/sensor/IR-filter combo. A lot of talk about filter designs but these curves illustrate some design choices.

So, what happens is that we essentially have a spectral response curve for each type  pixel (R, G and B). The illuminant has a similar curve and so does a point of the subject. We multiply the three curves for each pixel and get three signals. Those signal values go into the colour matrix.

This is of course a very coarse approximation, the only reason it works is that our human vision makes a very similar approximation.

Best regards
Erik

Before or after color correction matrix?

The difficulty that I have comprehending such spectral plots is that:
1) Any color that can be uniquely recorded can be made into any other color (using a 3-d lookup). There are good, practical reasons why this should not be done without limits.
2) Spectral selectivity that is far from the desired output would need "lots" of correction. Multiplying the 3 channels by a "nasty" 3x3 correction matrix would increase the likelihood of noise amplification viewed numerically. But a noise of "n" (as measured) might not map into a perceptual correlate very well.
3) A "bad" CFA colorwise might have other properties (avoid attenuating luminance, stable behaviour over time, ...) that makes it "good".

So it would seem that there is some room for optimizing the CFA for various tasks but I fail to see how the spectral selectivity can be analyzed directly without some accompanying color correction matrix/function in order to see the true trade-offs. At least for non-color-scientists.

-h

[torger]

I haven't really learnt to make any detailed conclusions from SSFs, more than those mentioned a couple of posts back. At some points I may look deeper into it again as it has some connection to designing camera profiles, which I do in my DCamProf project.

[Jack Hogan]

I have compared spectral plots from the Sony NEX5N and the Nikon D2X. They were very close, so I think Sony has the same recipe for high end pro and customer cameras.

Erik,

I understand that CFA recipes are closely guarded secrets - and Sony and Nikon's are not the same.

Jack

[torger]

I understand that CFA recipes are closely guarded secrets - and Sony and Nikon's are not the same.

It's hard to guard them, as they're very easy to measure if you have the equipment, which all camera manufacturers have for sure.

You can buy your own automated system here:
http://www.image-engineering.de/products/equipment/measurement-devices/587-camspecs

At €9750 it's not really a consumer toy, but it's small money for any larger company.

[ErikKaffehr]

Hi Jack,

What I have seen in these plots was that Nikon D3X and Sony NEX5M was very, very similar. But these were just a few plots of data I have found in in Anders Torger's references.

Would be interesting to now more!

I know that you have plotted some camera responses using very affordable means and I am very glad that you share findings, knowledge and experience!

Best regards
Erik

Erik,

I understand that CFA recipes are closely guarded secrets - and Sony and Nikon's are not the same.

Jack

[hjulenissen]

Spectral plots are recorded using a monochromator. So you expose frame after frame using near monochromatic light. A monochromator is an affordable device, but it needs to be calibrated by simultaneous reading using a photospectrometer and those are expensive.

The value of this plots it that they show the characteristics of the CFA/sensor/IR-filter combo. A lot of talk about filter designs but these curves illustrate some design choices.

I am sorry I should have chosen my words more clearly. I was not asking how the spectral plots are recorded (I am sure that they are accurate). I was asking if you were basing your conclusions on the spectral plots in isolation, or the spectral plots coupled with some correction.

I am guessing that the former is true. Then I question how one can suggest (even with disclaimers) that one CFA offers "subtle colour, more like Kodachrome than Velvia look" over another CFA.

My impression is that the physical color filtering mainly affects 1) The colors that cannot be distinguished (information is lost forever and cannot be reintroduced) and 2) Colors that can be distinguished, but with variable levels of effort needed (and secondary artifacts introduced). The two are probably the same thing, only distinguished by some system-dependent threshold.

My point is sort of similar to the Nikon D800 vs D800E discussion if we swap the color domain for spatial domain. By removing the OLPF, one gets more "selective" sensels (acutance). If such acutance is what the user wants, then there might be some advantage to doing it early in the recording chain (by removing the OLPF) instead of later. But to really know, one has to do a side-by-side, process both either "optimally" or "using the tools that some given user will use in practice", and see if and to what degree sharpening/deconvolution is able to equalize differences. I believe that Bart has done just such an exercise in the spatial domain.

One probably will find that in some cases, the differences are negligible (system linear errors are small enough that they can be practically compensated without banging your head against recording nonlinearity/noise/characterization errors), while in some cases the differences are significant. I imagine that the color filter may well be very "wrong", but as long as a nice transformation can make the result right, then the end user will be happy. A trivial example may be swapping the red and blur channels (very "wrong" colors, but easy to fix). A slightly less trivial example might be rotated primaries. "CMY" color filters may be very "wrong", but if the output can be transformed into some standard rgb representation through a nice linear matrix (or a full 3-d lookup) without any regions being "stretched too much" (noise amplification) or "compressed too much" (wasted recording precision), things could concievably be fine? (I don't know that CMY can actually be used, but I am suggesting it as an example).

-h

[Jack Hogan]

What I have seen in these plots was that Nikon D3X and Sony NEX5M was very, very similar. But these were just a few plots of data I have found in in Anders Torger's references.

Yes, Anders is the master and I must confess to not yet understanding color science, thus reading his and Jim Kasson's sites in silent adulation.

I do have many questions, though: for instance in one of the older papers there is a single plot with the spectral response of the CFAs of a fairly large number of DSLRs: differences appear minimal compared to the relatively huge transformations each curve will be subjected to on its way to being rendered into even just a neutral image.

Since the SSFs are so similar and the transformations so relatively huge, isn't it more likely that one camera's color response or 'discrimination' be more the result of the latter, rather than the former? 

Jack

[torger]

I'm much too humble to call myself a master on this :-). Despite writing some pretty advanced software dealing with SSFs I have certainly not fully understood the effect of the various shapes, to do that I would have to make much more experiments and comparisons than I've done.

What's true for sure though is that yes the resulting color in your image is mostly about the profile design. You just have to compare the color in say Capture One and Adobe Lightroom for the same file to see how large difference there is, and often when you apply a different camera's profile to your image you get pretty similar color, because SSFs are so similar these days.

In theory you can make cameras look exactly the same just by having a LUT that stretches all colors into the desired positions. However when you make a profile for generic use you don't want too strong non-linear corrections as that hurt gradients, thus you need to relax the LUT and then you come closer to the properties of the underlying SSF. It seems like the more saturated colors you compare, the easier it is to note differences related to the SSF.

I've attached identical shots made with a Leaf Aptus 75 (Dalsa) Hasselblad H4D-50 (Kodak) which has quite different SSFs. The profiles used where designed in a similar way, but not from this shot. The darker blue of the H4D-50 is probably partly due to a design difference rather than a SSF difference (the H4D-50 profile is made with an older version of DCamProf which made more realistic deep blues but more problematic in the extreme range). I think this is a fair visual demonstration that shows how large differences you can expect when you make a profile for generic use (that is with some LUT relaxation) with the same design goal for two different cameras with very different SSFs. The look is still pretty similar, but there are differences.

One difference that I think is related to the SSF is the red apple, more brownish of the Leaf and deeper saturated of the Hasselblad. The more realistic color (except the deep blue) is the Leaf as the Dalsa has a more neutral SSF. I shoot landscape, and to me those minor differences in color is acceptable so I generally don't worry at all about the SSFs. I would have preferred the Dalsa SSFs but I still upgraded to the Hasselblad as it had some other features I wanted more.

Note that if I would have designed both profiles simultaneously from the same source image and with the exact same design parameters the result would be even closer than this, and then the Kodak is "old school" with today's standards a pretty subjective SSF, so two cameras with modern sensors would be even closer still. So yes, I wouldn't put that much weight into SSFs. It is interesting though as "that special MF color" that some claim to see some think is partly due to SSF differences, and when the IQ250 was released it was stated in market material that it was one important factor. I'm not so sure that it is...

The old "MF truth" that Kodaks are more saturated than Dalsas has some merit though and is anchored in the SSFs, and I think it's that what we see in that red apple. But obviously if you design for the same target the difference will still be small.

[ErikKaffehr]

Hi,

One of the reason I plotted those data was that there is a lot of talk about DSLRs/CMOS having more permissive filters to provide higher ISO.

I also would think that colour profiles matter more than CFA designs.

What I have noticed that the curves for D2X (Early generation CCD based pro camera using a Sony sensor) virtually overlap NEX5N (Quite recent CMOS camera). So it seems that Sony uses the same CFA design for old CCD and present CMOS.

The other thing I noted is that the Hasselblad has almost no sensivity on red channel below 570 nm. The Phase One goes down to 540 nm. So that must mean that the Hassy back tested would see everything below 570 nm as green, blue or a mix of the two.

Comparing with film, Velvia also had a cut off on the cyan forming layer around 570 nm while Kodakchrome went much lower.

Sensitivity curves for different films and sensors are include below:

• Kodachrome 25
• Provia 100
• Velvia 50
• Phase One P1 (unspecified), Sony Nex5N, H2D (unspecfied)

I don't think a lot of conclusions can be drawn from these curves. What they don't show is that CMOS sensors are much different from CCD sensors. So, I would suggest that CCD sensors having better colour response is just another myth. Personally, I would guess that overlapping sensitivity curves are not a bad thing.

Best regards
Erik

Yes, Anders is the master and I must confess to not yet understanding color science, thus reading his and Jim Kasson's sites in silent adulation.

I do have many questions, though: for instance in one of the older papers there is a single plot with the spectral response of the CFAs of a fairly large number of DSLRs: differences appear minimal compared to the relatively huge transformations each curve will be subjected to on its way to being rendered into even just a neutral image.

Since the SSFs are so similar and the transformations so relatively huge, isn't it more likely that one camera's color response or 'discrimination' be more the result of the latter, rather than the former? 

Jack

[hjulenissen]

In theory you can make cameras look exactly the same just by having a LUT that stretches all colors into the desired positions. However when you make a profile for generic use you don't want too strong non-linear corrections as that hurt gradients, thus you need to relax the LUT and then you come closer to the properties of the underlying SSF. It seems like the more saturated colors you compare, the easier it is to note differences related to the SSF.

It is interesting that when using paper/printer profiles, we get to choose between "perceptual" and "relative", but no such thing when applying camera profiles.

For those who really need "accurate" colors (I am not one of them), would it not make sense to make a "color developer" that took into account camera vs lighting measurements, scene content and user preferences? If you image contains smooth gradations (e.g. sky) perhaps keeping them smooth is preferred over color accuracy. But if the image contains saturated patches of a single color (e.g. a red piece of clothing), perhaps it is better to render that patch "accurately" even if that means excessive noise in gradations that is not there in the scene anyways?

Or perhaps this is excessive engineering and everyone mess with their colors to achieve a subjective goal anyways.

-h