Is there a reason for the blue/cyan bias on the Sony A7RII images?

[BrianVS]

http://www.sony.net/Products/SC-HP/cx_news_archives/img/pdf/vol_65/imx081_091_111pq.pdf

This Sony publication compares a BSI with a FSI sensor.

The peaks for green and red are shifted towards shorter wavelengths, the blue peak is shifted a little to green. Overall, a "cold shift".

Add it all up, and you get some indication of a color shift in BSI sensors. Of course these curves are RGB, meaning they are with the CFA filter. If Sony published Spectral Response for monochrome sensors, it would be easier to see what is going on inherent to the difference in technology.

[Jack Hogan]

http://www.sony.net/Products/SC-HP/cx_news_archives/img/pdf/vol_65/imx081_091_111pq.pdf

This Sony publication compares a BSI with a FSI sensor.

The peaks for green and red are shifted towards shorter wavelengths, the blue peak is shifted a little to green. Overall, a "cold shift".

Add it all up, and you get some indication of a color shift in BSI sensors. Of course these curves are RGB, meaning they are with the CFA filter. If Sony published Spectral Response for monochrome sensors, it would be easier to see what is going on inherent to the difference in technology.

Brian, nice link, take my challenge above.  Which would you prefer for less 'shift', A) or B) ?  It's a trick question

Let's say the camera produces A) but we'd really like it to produce B).  In that case, how much of a difference do you think the tiny changes in Spectral Response in the PDF you linked are going to have on the color of the final image once the data makes it through all the processing necessary to display it?

Jack

[ErikKaffehr]

Hi,

I don't see significant differences…

Best regards
Erik

Brian, nice link, take my challenge above.  Which would you prefer for less 'shift', A) or B) ?  It's a trick question

Let's say the camera produces A) but we'd really like it to produce B).  In that case, how much of a difference do you think the tiny changes in Spectral Response in the PDF you linked are going to have on the color of the final image once the data makes it through all the processing necessary to display it?

Jack

[ErikKaffehr]

Hi,

My take is that 'A' would separate green, yellow and red better, but would not be very subtle. I guess 'B' would be more subtle. I also notice the blue sensivity on L channel in 'B'.

Best regards
Erik

You may very well be right, but until someone posts comparable Spectral Response (absolute QE in Kodak lingo) diagrams of the sensors in question as used in those two cameras all we've got to go on are the SMIs, which say the two cameras are comparable.  I think there is a lot of room for confusion and misunderstanding (especially on my part:-) where color is concerned.  For instance, in order to minimize the green discrimination problem you seem to be having with the a7RII in the woods, which of these two CFAs would you prefer?



C'mon Edmund, buy you a beer if you play

Jack

[BrianVS]

You may very well be right, but until someone posts comparable Spectral Response (absolute QE in Kodak lingo) diagrams of the sensors in question as used in those two cameras all we've got to go on are the SMIs, which say the two cameras are comparable.  I think there is a lot of room for confusion and misunderstanding (especially on my part:-) where color is concerned.  For instance, in order to minimize the green discrimination problem you seem to be having with the a7RII in the woods, which of these two CFAs would you prefer?



C'mon Edmund, buy you a beer if you play

Jack

I prefer "A". I don't like the double peak of the Red channel corrupting the Blue response and would not ever want a CFA with these dyes. It would be easy to start with "A" and produce the "B" response mathematically from the Raw file of "A". Most Blue and Green channels have a double peak, but the secondary peak is in the NIR region, and this can be filtered out using an IR blocking filter.

It is very frustrating to spend a lot of money on a camera and have the manufacturer hold back the actual spectral response data. I prefer my Kodak/OnSemi cameras as I know the spectral response from the data sheets.

As far as the BSI vs FSI plots shown in the Sony link, shifting the peaks by 20nm is going to produce a difference. It can always be corrected in the color balance.

To add: The Red curve in "B" looks similar to the dye used in Canon DSLR's. I would not have expected to see a Red dye with such a strong secondary peak in the Blue region to be used. Why did Canon select this dye?

http://www.maxmax.com/spectral_response.htm


The Red dye used by Kodak has a secondary peak, but it is in the UV and is not an issue.

http://www.onsemi.com/pub_link/Collateral/KAF-18500-D.PDF

[AlterEgo]

Add it all up, and you get some indication of a color shift in BSI sensors. Of course these curves are RGB, meaning they are with the CFA filter.

we can see a color shift in a specific pair of different sensors (small sensels vs 42mp FF) with CFA (so you can't actually say whether, and which, the shift was because of CFA or not), w/o IR/UV cut (and actual camera has it) filter and w/o color transform (by raw converter using a specific camera profile)...

[BrianVS]

we can see a color shift in a specific pair of different sensors (small sensels vs 42mp FF) with CFA (so you can't actually say whether, and which, the shift was because of CFA or not), w/o IR/UV cut (and actual camera has it) filter and w/o color transform (by raw converter using a specific camera profile)...

Obviously the spectral response shown in the Sony publication are done without using an IR blocking filter, the graphs just do not go all the way out to 1.1uM.

Spectral response curves are independent of color transforms and raw converter.

If spectral resonse curves were published for monochrome sensors, then we could differentiate what the effects of the BSI vs FSI technology was.

Until then- yes, BSI sensors are more efficient in absorbing light at shorter wavelengths than are FSI sensors. Most manufacturers are not going to throw away the QE gains in the CFA.

[Jim Kasson]

I prefer "A". I don't like the double peak of the Red channel corrupting the Blue response and would not ever want a CFA with these dyes. It would be easy to start with "A" and produce the "B" response mathematically from the Raw file of "A". Most Blue and Green channels have a double peak, but the secondary peak is in the NIR region, and this can be filtered out using an IR blocking filter.

I think the opposite. The short lambda peak in the "red" channel is the only way to get the correct response to spectral violet.

I remind you that the rho cone cells have two peaks.

Jim

[Jim Kasson]

Erik, I've seen this bias in Lr with AWB engaged. However, if I set the WB to that of the lights (love my Wescott multi-WB LED panels) I get good WB. And, of course, WB to a gray card is fine.

http://blog.kasson.com/?p=12553

Jim

[AlterEgo]

Until then- yes, BSI sensors are more efficient in absorbing light at shorter wavelengths than are FSI sensors.

with this narrow fact there is no argument.

[AlterEgo]

Erik, I've seen this bias in Lr with AWB engaged. However, if I set the WB to that of the lights (love my Wescott multi-WB LED panels) I get good WB. And, of course, WB to a gray card is fine.

http://blog.kasson.com/?p=12553

Jim

and then again - in LR/ACR WB depends on dcp profile (as it uses the matrices there)... make a different profile (not resuising Adobe matrices) and you will get a different result of WB operation (not only what is displayed as K/tint in UI)...

[Jim Kasson]

and then again - in LR/ACR WB depends on dcp profile (as it uses the matrices there)... make a different profile (not resuising Adobe matrices) and you will get a different result of WB operation (not only what is displayed as K/tint in UI)...

True enough.

However, I got the same results with ASP"

http://blog.kasson.com/?p=12524

and C1 generic:

http://blog.kasson.com/?p=12573

Jim

[eronald]

Jack, me not play.
Once you have covered the sensor with a CFA and IR filter, there are three independent issues:
- Luther Ives which determines how the rendering resembles human visison.
- DR after the CFA.
- Filter orthogonality which tells you how ugly or pretty the matrix is, and how good your separation is in practical terms, once you have lost DR to wb and ISO push.

My feeling is that the new Sony slightly suffers from both.

Edmund

True enough.

However, I got the same results with ASP"

http://blog.kasson.com/?p=12524

and C1 generic:

http://blog.kasson.com/?p=12573

Jim

[Jack Hogan]

I prefer "A". I don't like the double peak of the Red channel corrupting the Blue response and would not ever want a CFA with these dyes. It would be easy to start with "A" and produce the "B" response mathematically from the Raw file of "A". Most Blue and Green channels have a double peak, but the secondary peak is in the NIR region, and this can be filtered out using an IR blocking filter.

It is very frustrating to spend a lot of money on a camera and have the manufacturer hold back the actual spectral response data. I prefer my Kodak/OnSemi cameras as I know the spectral response from the data sheets.

As far as the BSI vs FSI plots shown in the Sony link, shifting the peaks by 20nm is going to produce a difference. It can always be corrected in the color balance.

To add: The Red curve in "B" looks similar to the dye used in Canon DSLR's. I would not have expected to see a Red dye with such a strong secondary peak in the Blue region to be used. Why did Canon select this dye?

Perhaps Canon (and Nikon) did because B) is one recent estimate of the photopic response of the cones in the average human eye, as Jim correctly hinted to.  Did I forget to mention it was a trick question?

Anyways I am definitely out of my depth here and glad to see that real color scientists have joined the party who can jump in to save me when I start sinking : the point I was trying to make is that B) could represent one ideal set of CFA recipes while A) is quite far from that ideal - although on the other hand A) is quite typical of CFAs used in current digital still cameras.  The CFA in B) could in theory be used as-is to capture excellent color information, while data collected with CFA A) will require massive transformations before it can be used to display an approximation of the color information from the scene as perceived by the average human.

So given that A) is so far off and requires such mathematically intensive manipulation just to get approximately pleasing color out of it, what will such tiny differences as those shown in the Sony spectral response graphs above mean practically? Very little says Erik.  The differences look to me like they could almost be classified as measurement error, so I concur.  Don't forget, if two curves are similar (and these seem so to me) what gets recorded in the raw data is proportional to the area under each curve.  Once appropriate filters for general photography are in place, how different are those areas?  Once white balanced?  Once subjected to color transformations?  My guess is that there is a lot more play elsewhere in the system which will mask those differences, making them virtually immaterial.

Jack

[Jack Hogan]

Jack, me not play.
Once you have covered the sensor with a CFA and IR filter, there are three independent issues:
- Luther Ives which determines how the rendering resembles human vision.
- DR after the CFA.
- Filter orthogonality which tells you how ugly or pretty the matrix is, and how good your separation is in practical terms, once you have lost DR to wb and ISO push.

My feeling is that the new Sony slightly suffers from both.

Edmund

All work and no play makes Jack a dull boy   Ok, so on the work side and for my edification:  since we know that the a7RII has good - in fact class-leading - DR after the CFA at 'ISO push' compared to other current advanced DSCs, does it just boil down to the orthogonality of the CFA curves?  How does one define orthogonality of the curves?  Assuming their shapes are pretty typical would it be related to the position and distance of the three peaks, and their half power bandwidths?  If so and since we are at it, how does the human visual system discriminate green so well with the apparent lack of orthogonality in B)?  Any help by anyone would be appreciated.

Jack

[BrianVS]

Perhaps Canon (and Nikon) did because B) is one recent estimate of the photopic response of the cones in the average human eye, as Jim correctly hinted to.  Did I forget to mention it was a trick question?

Jack

The eye declares color based on a non-additive process, in which "very simplified explanation" tests which cones respond and which ones don't. Certain cells are inhibited from responding in the presence of certain colors. Like positive and negative voting by groups of cells responding to bands of spectra. I saw a spectral curve of the human eye that was stated to be an RGB response that had the double peak in Red while looking for the response curve that matched "B". Most response curves for the eye do not show this second peak. Overall sensitivity of the eye is much different, with blue response being low. One article stated that only 2% of the cones of the eye are "Blue" cones. This is one reason for using a yellow filter when shooting the sky with B&W.

Having the double peak with a camera using additive color- it's blue contamination in the red channel, like IR contamination on Blue and Green if you don't block IR. I use a deep yellow filter on the M Monochrom for most shots. I like to use an Orange filter (block blue and leave the blue channel sensitive to IR only) on the M8 for color-Infrared, then boost the blue channel and swap Blue and Red in the DNG file header.

[Jack Hogan]

The eye declares color based on a non-additive process, in which "very simplified explanation" tests which cones respond and which ones don't. Certain cells are inhibited from responding in the presence of certain colors. Like positive and negative voting by groups of cells responding to bands of spectra. I saw a spectral curve of the human eye that was stated to be an RGB response that had the double peak in Red. Most response curves for the eye do not show this second peak. Overall sensitivity of the eye is much different, with blue response being low. This is one reason for using a yellow filter when shooting the sky with B&W.

Interesting, Brian.  Upon further investigation it seems that the detailed shape of the curves depend on the stage at which one looks at them (absorption, fundamental, intermediate, final).  I see that in Wyszeki and Stiles a mild second 'red' cone bump tends to be there in graphs where the y-axis represents 'Spectral Absorption Coefficient' or 'Relative Absorptance'.  The y-axis of the bottom graph I posted was labeled 'spectral response' (although in spite of that upon second glance its curves are starting to look suspiciously like color matching functions for the 1931 standard observer).  Here are the Stockman & Sharpe cone fundamentals normalized to 1 at each peak from the Color and Vision Research Lab at UCSD, the vertical axis being relative energy.  I would assume that the un-normalized version of these curves would need to be weighted by the photopic luminosity curve.

Having the double peak with a camera using additive color- it's blue contamination in the red channel, like IR contamination on Blue and Green if you don't block IR. One article stated that only 2% of the cones of the eye are "Blue" cones. I use a deep yellow filter on the M Monochrom for most shots. I like to use an Orange filter on the M8 for color-Infrared, then boost the blue channel and swap Blue and Red in the DNG file header.

What strikes me in these graphs is how narrow and separated (orthogonal?) blue is compared to green-red.  How much do green cones contribute to our perception of color in normal daylight vision?
And how spread out and close to green the blue curve is in CFAs I have seen measured for a few Nikon and Canon cameras.  Is the 'green discrimination' issue mainly a function of the intrusive blue channel?

Jack

[Jim Kasson]

Interesting, Brian.  Upon further investigation it seems that the detailed shape of the curves depend on the stage at which one looks at them (absorption, fundamental, intermediate, final).  I see that in Wyszeki and Stiles a mild second 'red' cone bump tends to be there in graphs where the y-axis represents 'Spectral Absorption Coefficient' or 'Relative Absorptance'.  The y-axis of the bottom graph I posted was labeled 'spectral response' (although in spite of that upon second glance its curves are starting to look suspiciously like color matching functions for the 1931 standard observer).  Here are the Stockman & Sharpe cone fundamentals normalized to 1 at each peak from the Color and Vision Research Lab at UCSD, the vertical axis being relative energy.  I would assume that the un-normalized version of these curves would need to be weighted by the photopic luminosity curve.



What strikes me in these graphs is how narrow and separated (orthogonal?) blue is compared to green-red.  How much do green cones contribute to our perception of color in normal daylight vision?
And how spread out and close to green the blue curve is in CFAs I have seen measured for a few Nikon and Canon cameras.  Is the 'green discrimination' issue mainly a function of the intrusive blue channel?

Whenever someone quotes W&S, you know they're a force to be reckoned with. Well played, Jack.

I remember a meeting with a Kodak Research color scientist at Almaden Research Center. First thing he dds was pull a copy of W&S out of his briefcase and set it on the table. Then we talked.

A little background:

http://blog.kasson.com/?p=12462

The big reason that cameras don't use overlapping red and green filters like the rho and gamma cones is SNR. Takes a lot of subtraction to get from there to, say, Adobe RGB.

Jim

[Jim Kasson]

Interesting, Brian.  Upon further investigation it seems that the detailed shape of the curves depend on the stage at which one looks at them (absorption, fundamental, intermediate, final).  I see that in Wyszeki and Stiles a mild second 'red' cone bump tends to be there in graphs where the y-axis represents 'Spectral Absorption Coefficient' or 'Relative Absorptance'.  The y-axis of the bottom graph I posted was labeled 'spectral response' (although in spite of that upon second glance its curves are starting to look suspiciously like color matching functions for the 1931 standard observer).  Here are the Stockman & Sharpe cone fundamentals normalized to 1 at each peak from the Color and Vision Research Lab at UCSD, the vertical axis being relative energy.  I would assume that the un-normalized version of these curves would need to be weighted by the photopic luminosity curve.

These curves would look quite different if weighted by the population of each type of cone.

Jim

[Jim Kasson]

What strikes me in these graphs is how narrow and separated (orthogonal?) blue is compared to green-red.  How much do green cones contribute to our perception of color in normal daylight vision?

The gamma cones contribute a lot to our perception of color. They are one axis of three. The beta cones contribute little to luminance, but a great deal to the opponent color (red/green and blue/yellow) results. We wouldn't have a red/green axis without the gamma cells.

BTW, with a lens that has lousy LoCA, it makes a lot of sense for the eye to put most of the luminance sensing in one part of the spectrum.

Jim