p.1 #1 · Bayer masks and DxO color depth measurements
There is rare, but consistent information on photography forums suggesting, that some camera makers are tuning their sensors for high ISO performance (Canon, Nikon?) and some for better ability to separate hues (MF backs, Leica, Sony Alpha?). This tuning is apparently achieved by different spectral characteristics of R, G and B filters in the CFA array. There is also the DxO Mark web site, which makes available their "color depth" data, which according to their description measures the same thing. Their results, however, do not indicate any special hue sensitivity/color depth of Leica or Alpha cameras. In fact, Nikon D800, for example, scores higher. I spent hours on the internet trying to find some explanation for this discrepancy, but I still don't have any satisfactory answer. Can anybody help me understand what I am missing?
I realize that DxO Mark results spun many heated discussions before, but these discussions usually revolve around their aggregate scores, not individual detailed measurements. I am concerned in one particular individual measurement, that looks reasonable, and I hope this question is not going to generate too much negative reaction.
p.1 #2 · Bayer masks and DxO color depth measurements
You'll get the information you're looking for in DxO's "Color Response" tab for the sensor results (not available in the comparison view btw). Hopefully theSuede will come by to weight in but here's what I think I've learned from his and iliah borg's posts...
The Sensitivity Metamerism Index describes how well the sensor avoids metamers, which are situations where different sets of "real-world" colors are mistakenly interpreted as the same color by the sensor. The higher the SMI number, the fewer meters it has (better).
The raw -> sRGB matrix describes how much digital gain must be applied to each raw channel (RGB) in order to produce an sRGB primary color. The more selective a sensor is in distinguishing color between the RGB filters, the less gain that needs to be supplied to produce a given color, which translates to less chroma noise esp. in PP situations where saturation is boosted. The way to read DxO's matrix per theSuede is diagonally, top-left to bottom-right, taking the three numbers on that diagonal to see the color purity per channel. Making the sensor's RGB CFA filter more transmissive (less selective) means better High ISO performance but at the expense of increase color noise or reduced saturation...this is why DxO includes color depth into their overall High ISO score calculation, and why Canon sensor's results are lower than others even though their noise and DR is about the same as equivalent sensors.
DxO shows results for two lighting temperatures, CIE-D50 (daylight) and CIE-A (tungsten). The difference between the two is significant when evaluating how the sensor will perform for your intended lighting conditions.
p.1 #3 · Bayer masks and DxO color depth measurements
Making the sensor's RGB CFA filter more transmissive (less selective) means better High ISO performance but at the expense of increase color noise or reduced saturation...this is why DxO includes color depth into their overall High ISO score calculation, and why Canon sensor's results are lower than others even though their noise and DR is about the same as equivalent sensors.
Thank you. This was my understanding, too, yet Nikon sensors score in the color depth index higher than Alpha and Leica. That's the part that I don't get.
p.1 #5 · Bayer masks and DxO color depth measurements
Mirek Elsner wrote:
Thank you. This was my understanding, too, yet Nikon sensors score in the color depth index higher than Alpha and Leica. That's the part that I don't get.
p.1 #6 · Bayer masks and DxO color depth measurements
Well, that is a different measurement, relating to which colours cannot be differentiated. The colour depth is more strongly related to the quality of the CFA, AFAIK.
p.1 #7 · Bayer masks and DxO color depth measurements
Mirek Elsner wrote:
Thank you. This was my understanding, too, yet Nikon sensors score in the color depth index higher than Alpha and Leica. That's the part that I don't get.
Because their color depth is a measurement of per-channel (color) noise, which relates to both the base efficiency of the sensor (Quantum Efficiency) and the per-channel efficiency/separation. All contemporary Nikon/Sony sensors are near the top for QE and have reasonably good color separation, which is why they score so well for color depth.
p.1 #8 · Bayer masks and DxO color depth measurements
Yep, SMI is that primary part of "how good colors can be".
Next important thing (for me) is if ADC is 12-bit or 14-bit (tho Sony is very effective even with 12).
Bit depth should show how good will be transitions and how much "colors" will be visible (at least thats what I understand from their explanation). I actually asked them long time ago about this, but didnt get any answer..
Making CFA (its far from pure RGB in case of Canon) less selective will end in colors not being separated enough. We can imagine shot of forest - with good CFA, there will be a lot of various hues of green, with bad CFA there will be mostly one type of green, that will vary only in intensity, not hue. Simply one green, darker or lighter.
Some of these low CFA cams have usually various tints to whole image (probably possible to "fix" via good color profile).
p.1 #9 · Bayer masks and DxO color depth measurements
Mirek Elsner wrote:
There is rare, but consistent information on photography forums suggesting, that some camera makers are tuning their sensors for high ISO performance (Canon, Nikon?) and some for better ability to separate hues (MF backs, Leica, Sony Alpha?). This tuning is apparently achieved by different spectral characteristics of R, G and B filters in the CFA array. There is also the DxO Mark web site, which makes available their "color depth" data, which according to their description measures the same thing. Their results, however, do not indicate any special hue sensitivity/color depth of Leica or Alpha cameras. In fact, Nikon D800, for example, scores higher. I spent hours on the internet trying to find some explanation for this discrepancy, but I still don't have any satisfactory answer. Can anybody help me understand what I am missing?
I realize that DxO Mark results spun many heated discussions before, but these discussions usually revolve around their aggregate scores, not individual detailed measurements. I am concerned in one particular individual measurement, that looks reasonable, and I hope this question is not going to generate too much negative reaction. ...Show more →
p.1 #12 · Bayer masks and DxO color depth measurements
I posted this on Get DPI, too, but here is some of thesuede's explanation. He's the guy to talk to about this. It doesn't look like DxO Mark is the best place to determine CFA quality: link
Comparing the DxO "portrait" (color depth) scores, the MF digital backs consistently come out at the very top for their introduction date. The "metamerism index" indicates how well the distinction of colors lines up with that of the human eye; the MF backs don't seem so good at that. However, their high "color depth" indicates that they can resolve very fine color differences --- just not exactly the same color differences that the human eye would see. I would guess that this quality turns out to be more important to MF back users; DxOMark downplays the importance of MI to actual sensor quality:
In practice, the SMI for DSLRs ranges between 75 and 85, and is not very discriminating. It is different for low-end cameras (such as camera phones), which typically have a SMI of about 40. For this reason, we give this measurement as an indication but do not integrate it in DxO Mark.
p.1 #16 · Bayer masks and DxO color depth measurements
Mirek Elsner wrote:
It was my impression that what theSuede says resonates very well with the detailed color depth analysis on DxO Mark. But I probably miss something.
It would be great if he weighed in, but to quote him from above, "You can only guess at what the CFA filter spectrum looks like from the DxO figures - but they do plot complete spectrals at some places not directly connected to the "DxO-mark" part of the site. I don't know if all of them are open to the general public."
p.1 #17 · Bayer masks and DxO color depth measurements
Comparing the DxO "portrait" (color depth) scores, the MF digital backs consistently come out at the very top for their introduction date. The "metamerism index" indicates how well the distinction of colors lines up with that of the human eye; the MF backs don't seem so good at that.
The inverse relationship of metamerism and color depth is in line with what theSuede says, I think. But Leica M9 and Alpha did not come out stellar. I am trying to find some evidence that the tradeof of worse bad ISO is worth it. I know people like colors from Alpha and Leica, but I am looking for something solid that shows that there is really better separation.
p.1 #18 · Bayer masks and DxO color depth measurements
In some of these cases (maybe Leica) I think that the default profiles and processing are just more pleasing. This was the angle that the film makers took. Not accurate, but pleasing. I would be happy if Nikon moved in that direction too, sometimes getting the colours I want is a bit of work.
p.1 #19 · Bayer masks and DxO color depth measurements
douglasf13 wrote:
It would be great if he weighed in, but to quote him from above, "You can only guess at what the CFA filter spectrum looks like from the DxO figures - but they do plot complete spectrals at some places not directly connected to the "DxO-mark" part of the site. I don't know if all of them are open to the general public."
Right. Seeing the actual spectrum would eliminate the speculations which cameras use less transmissive arrays. Now we only speculate that Alpha and Leica have less transmissive arrays and should give better separation.
p.1 #20 · Bayer masks and DxO color depth measurements
Mirek Elsner wrote:
Right. Seeing the actual spectrum would eliminate the speculations which cameras use less transmissive arrays. Now we only speculate that Alpha and Leica have less transmissive arrays and should give better separation.
Based on the DxO data the M9 doesn't appear to have an especially good color profile. And its relatively poor High ISO performance is due mostly due its sensor's low quantum efficiency rather than the density of its CFA.
