| LR6 | Capture One | |
| P45+ | (http://echophoto.dnsalias.net/ekr/Articles/OLS_OnColor/SimpleCase/20150107-CF046070_AdobeStandard_vsmall.jpg) | (http://echophoto.dnsalias.net/ekr/Articles/OLS_OnColor/SimpleCase/20150107-CF046070_C1_vsmall.jpg) |
| Sony A99 SLT | (http://echophoto.dnsalias.net/ekr/Articles/OLS_OnColor/SimpleCase/20150107-_DSC6397_AdobeStandard_vsmall.jpg) | (http://echophoto.dnsalias.net/ekr/Articles/OLS_OnColor/SimpleCase/20150107-_DSC6397_C1_vsmall.jpg) |
off topic, but ...
Iliah Borg suggested elsewhere the following to read = http://www.aphesa.com/downloads/download2.php?id=1 which illustrates that "smooth" SSF/CMF curves you can find (mostly, not all) around are not exactly what is happening (and rather illustrate some sloppy work on behalf of measurebators) - here is his measurement of the unknown sensor (not normalized between channels)
(http://s26.postimg.org/ga85dvmg7/Sensor_Sawtooth_copy.jpg)
The image you linked to is not visible. Could you fix that link, if possible?
not sure why, in my browser (FFox) it is OK = http://s26.postimg.org/ga85dvmg7/Sensor_Sawtooth_copy.jpg
My intention with plots was to show similarities and differences between sensors and for that purpose I think they are good enough.the devil is in the details... I suggested, rather made a guess, (elsewhere where it was posted) that as we are dealing with integrating anyways then more precise sawtooth does not add any "noticeable" precision vs smooth graphs (interpolated from 10nm steps), but then he replied then methameric errors will be very noticeable... so even very similar graphs for the same camera (one measured casually and one with efforts to uncover this effect) can really be noticeably different, no ?
coca cola commercials?
My opinion is that colour reproduction depends much more on colour profiles than on camera sensors.
Oh, a quick comment on the low SMI score of the P45+ shown in Tim's article. Kodak liked to make subjective colors and their sensors have quite saturated native colors, and indeed a matrix-only profile can't match the CC24 that well -- hence a poor score. Another reason to a relatively low score is that the Kodak sensor is noisy compared to a modern Sony sensor.
A non-linear profile (LUT) can make a quite good match with the CC24 though if you want to, but here's the interesting part -- color is subjective and if you decide to work with the Kodak subjective colors you can make a clean robust profile with a subjective look which you actually may like. So color is not necessarily bad due to the low score, in this case it's a strong indication that the color is subjective and not well-suited to make reproduction style color, but may produce a pleasing look.
How well does SMI correlate to "color separation" or "color accuracy"? Uhmm... not so much I'd say. You must factor in the color profile. It's true that a high SMI score is a good basis to build a color profile with any look you want both subjective and neutral, while a low score locks you in with some subjectivity. If you pick any default color profile in Capture One or Lightroom or whatever there's a lot of subjectivity built-in already, which is shown in the initial post. With Lightroom there's also very different profile designs for different cameras so it's hard to compare different cameras even within Lightroom. That may be the case with Capture One too, I don't know. In any case I haven't seen any indications that you would be able to "see through" the color profile and make conclusions about how the native raw colors are underneath.
Photographers that worry about color should really make their own profiles to take control over the situation. Unfortunately there's not so much software out there to do that. That was the reason I wrote DCamProf but I can't say it's a software "for the masses", it's a bit too complicated to use for that.
To get good high ISO you want to keep as much light as possible, that is ideally the camera would be monochrome.
Color separation is a complex area though. A good basis for good color separation is a low noise sensor, then you can separate colors even if the channels differ very little. That is you can have wide overlaps and let through lot of light in all filters and still be able to separate colors.
With a noisy sensor it's better to have more difference between color filters so the channels differ more (you get more saturated colors natively). In extreme cases you don't have overlap at all which I think was the case for some older prismatic TV cameras in the analog days. This indeed gives very sharp separation, but due to the lack of overlap there are colors that will be registered to be the same that would be seen as different with overlapping filters.
All(?) modern sensors today have wide overlaps and and quite low saturation colors natively. This allows for registering lots of colors, and improves ISO as more photons are captured, but also put requirement on the sensor to be low noise so we can through the color profile add back saturation and visibly separate colors.
When working with DCamProf I noted that many cameras, specifically Sony, had a high sensitivity on the blue channel. This made it very complicated to make realistic blue tones in daylight (daylight has high blue content in itself). The solution is to render blues lighter than realistic, which indeed is a popular "look" which many color profiles implement in way way or another, regardless if the sensor requires it or not. I felt that Canon's color filters where better balanced to produce realistic colors though.
The DxOmark SMI index is a measure how well the ColorChecker 24 color can be matched with a matrix-only profile. A disadvantage of that test is that it doesn't say anything on how the cameras react to "extreme" colors (very high saturation colors), so the Sonys fair very well (much thanks to their low noise) although I personally prefer the response I've seen in Canon cameras as their easier to manage in the extreme range.
Are there tradeoffs ISO vs color separation? Maybe/probably, but I don't really know how they are done. A guess is that the high blue channel sensitivity in the Sony is a high ISO thing, plus that they realized that most subjectively like lighter-than-realistic deep blues anyway. That doesn't really hurt color separation though, just the possibility to make realistic colors in cool light. You could make the color filters be more similar in the overlaps to let through more light, and increase the similarity more than what is suitable for an ideal color separation tradeoff.
With DCamProf I made a color separation diagram which can be used to make comparisons how well a camera can separate colors compared to the eye: http://www.ludd.ltu.se/~torger/dcamprof.html#ssf_csep
However one needs to factor in the sensor noise to make really good sense of that.
If we talk about the "medium format magic colors" or any other camera's supposedly magic color I'd say that it's almost only about well-designed subjective color profiles that fits the taste of their users. Any sensor today of reasonable size in good lighting condition have sufficient color separation to design almost any type of look you want, at least that's the indications I've got from my work, so for any ISO sacrifice made it doesn't seem to be large. Going from the camera's raw colors to a robust color profile is not an easy thing though. One may think it's only about shooting a chart and matching, which in the case of reproduction work it may be, but for all-around photography you need to apply contrast, a "film curve", and that will modulate colors as a side effect (as our brain's color perception is tightly linked to contrast). That's why matrix profiles designed for a linear curve works so bad with contrast applied, oh well bad if you ask me, some like the result anyway and it's much about taste.
A well designed color profile is designed to match the film curve it's intended to be used with. There are really no established color science models to do this so it's up to the profile designer to invent their own models, and the manufacturers have done so, and so have I with DCamProf. We all end up with different results of course, as we have different design targets and tastes.
Hi Anders,
Thanks for sharing your experience!
It is interesting to hear that you find the Canon approach a good one, although I don't feel it is a great surprise.
With the blue sensitivity on the Sony sensors, that has not been obvious to me. Can you also see it on Phase One (IQ-250) and Nikon that also use Sony sensors?
My guess from looking at the spectral plots was that some sensors are more balanced for incandescent light, so they have high sensitivity for blues. It makes a lot of sense for average shooting that is a mix of different colour temperatures.
Best regards
Erik
My opinion is that colour reproduction depends much more on colour profiles than on camera sensors.
Iliah Borg suggested elsewhere the following to read = http://www.aphesa.com/downloads/download2.php?id=1 which illustrates that "smooth" SSF/CMF curves you can find (mostly, not all) around are not exactly what is happening (and rather illustrate some sloppy work on behalf of measurebators) - here is his measurement of the unknown sensor (not normalized between channels)
The correct bluish purple is this (measured with a spectrometer):...
Green varies a lot more: ...
Such large oscillations would clearly be seen in SSF dataand he provided his measurements with such sawtooth, granted he has more elaborate setup that allows him to sample finer than 10nm even with monochromator set for 10nm (as you can also measure the light going into monochromator with finer than 10nm steps with spectrophotometer)...
for instance in the references Erik linkedif this was about RIT graphs those were quick proof of concept measurements w/o any attempt to calculate finer than 10nm data (and they had spectroradiometer better than some consumer level spectrophotomer like i1Pro-something) and you can see that they even keep the lens on the camera, instead of mounting camera to monochromator and measuring the transmission lens separately.
Erik, that's interesting. May I ask how you produced the 'correct' colors?
Jack
and some lenses are
I still don't think this is representative of current CFA sensors.why ?
why ?Because it's the only one of many, many that looks like that.
Because it's the only one of many, many that looks like that.well, I read the description of how he was measuring - it sounds way better than 20 graphs from that web page of Jinwei Gu (who was not exactly measuring for the ultimate precision)... but by all means you can engage him into a small talk 8)