Foveon vs CFA sensor daylight spectra - Part 2

[Stillton]

Part 1 - Foveon colors through a prism of a spectroscope DP3M: Sigma Camera Talk Forum: Digital Photography Review (dpreview.com)

After much f**ing around, I finally realized that I CAN take images of spectra from my spectroscope directly - by putting its ocular against the lens of a camera. That way the images come out a bit clearer and more contrast-y.
Anyway, here are the results:





Note1: vertical lines are not defects/artifacts but rather absorption lines of various elements in the atmosphere. These were super handy as I used them to align the results.
Note2: iPhone8 image has excellent contrast, because its lens was smaller than the ocular allowing for a nice optical seal.
This was not the case with my D5000 or DP3M, so some light was still getting into the lens and reducing contrast.

 

[SigmaChrome]

Part 1 - Foveon colors through a prism of a spectroscope DP3M: Sigma Camera Talk Forum: Digital Photography Review (dpreview.com)
After much f**ing around, I finally realized that I CAN take images of spectra from my spectroscope directly - by putting its ocular against the lens of a camera. That way the images come out a bit clearer and more contrast-y.
Anyway, here are the results:





Note1: vertical lines are not defects/artifacts but rather absorption lines of various elements in the atmosphere. These were super handy as I used them to align the results.
Note2: iPhone8 image has excellent contrast, because its lens was smaller than the ocular allowing for a nice optical seal.
This was not the case with my D5000 or DP3M, so some light was still getting into the lens and reducing contrast.

I find it fascinating that the Foveon sensor is the only one that seem to show any violet at all. Love to know why that is the case.

--
Regards,
Vitée
Capture all the light and colour!

Galleries by Vitee

www.pbase.com

 

[xpatUSA]

Part 1 - Foveon colors through a prism of a spectroscope DP3M: Sigma Camera Talk Forum: Digital Photography Review (dpreview.com)
After much f**ing around, I finally realized that I CAN take images of spectra from my spectroscope directly - by putting its ocular against the lens of a camera. That way the images come out a bit clearer and more contrast-y.
Anyway, here are the results:





Note1: vertical lines are not defects/artifacts but rather absorption lines of various elements in the atmosphere. These were super handy as I used them to align the results.
Note2: iPhone8 image has excellent contrast, because its lens was smaller than the ocular allowing for a nice optical seal.
This was not the case with my D5000 or DP3M, so some light was still getting into the lens and reducing contrast.

Another excellent presentation !!

I find it fascinating that the Foveon sensor is the only one that seem to show any violet at all. Love to know why that is the case.

Violet light spans ~380-450 nanometers

https://en.wikipedia.org/wiki/Violet_light

Also see here for Foveon vs Bayer:

http://kronometric.org/phot/sensor/fov/2002 Color Paper - Allen Rush.pdf

or here:

http://kronometric.org/phot/sensor/fov/CIC10_Lyon_Hubel_FINAL.pdf

 

[Stillton]

I find it fascinating that the Foveon sensor is the only one that seem to show any violet at all. Love to know why that is the case.

I have the same question. Foveon even advertised it in one of their famous images.



Foveon is top most, and 2nd - 4th are competitors, I think.

Interestingly that D5000 D2X MODE 3 sort of has some violets too, but obviously to a much lesser extent than the Foveon.

 

[Stillton]

Another excellent presentation !!

Thank you!

Violet light spans ~380-450 nanometers

What is interesting is that iphone has a bit wider spectral range than both Foveon and D5000. Yet Apple did not choose to resolve those as purple/violet... instead opting for deep blue.

You can see that those blues dont really change color in the entire "blue" spectral subrange on iPhone spectrum.

 

[xpatUSA]

I find it fascinating that the Foveon sensor is the only one that seem to show any violet at all. Love to know why that is the case.

I have the same question.

One way would be to start with the QE curves for a given Sigma/Foveon model then apply the appropriate 3x3 matrices (found in X3F meta-data) to convert to RGB.

Fairly clear for the early sensors - less so for Merrills and Quattros ...

... easiest being for the SD9/SD10 which converts cam to XYZ and stays in XYZ for the WB adjustments, after which the conversion matrix from XYZ to RGB is well-known in the literature.

 

[xpatUSA]

Another excellent presentation !!

Thank you!

Violet light spans ~380-450 nanometers

What is interesting is that iphone has a bit wider spectral range than both Foveon and D5000. Yet Apple did not choose to resolve those as purple/violet... instead opting for deep blue.
You can see that those blues dont really change color in the entire "blue" spectral subrange on iPhone spectrum.

Some time ago I found a calculated spectrum that claimed to give the best possible rendering of an equal-energy spectrum (so not Daylight) on an sRGB monitor. The link is long gone off my system but here is the image:


Interesting to compare it with the various camera spectra to date in this thread.

:-D

 

[Stillton]

Another excellent presentation !!

Thank you!

Violet light spans ~380-450 nanometers

What is interesting is that iphone has a bit wider spectral range than both Foveon and D5000. Yet Apple did not choose to resolve those as purple/violet... instead opting for deep blue.
You can see that those blues dont really change color in the entire "blue" spectral subrange on iPhone spectrum.

Some time ago I found a calculated spectrum that claimed to give the best possible rendering of an equal-energy spectrum (so not Daylight) on an sRGB monitor. The link is long gone off my system but here is the image:


Interesting to compare it with the various camera spectra to date in this thread.

:-D

Interesting that blue band is super narrow compared to violets, while the majority of modern CFA cameras do the opposite - lots of blues and almost no violet.

 

[xpatUSA]

Another excellent presentation !!

Thank you!

Violet light spans ~380-450 nanometers

What is interesting is that iphone has a bit wider spectral range than both Foveon and D5000. Yet Apple did not choose to resolve those as purple/violet... instead opting for deep blue.
You can see that those blues dont really change color in the entire "blue" spectral subrange on iPhone spectrum.

Some time ago I found a calculated spectrum that claimed to give the best possible rendering of an equal-energy spectrum (so not Daylight) on an sRGB monitor. The link is long gone off my system but here is the image:


Interesting to compare it with the various camera spectra to date in this thread.

:-D

Interesting that blue band is super narrow compared to violets, while the majority of modern CFA cameras do the opposite - lots of blues and almost no violet.

Right. I just finished looking at the iPhone blues - extracted as Hues in the GIMP:


Almost monochromatic !

One can only speculate if there is any advantage to that rendering ...

 

[DavidWright2010]

I opened the image of spectra with ImageJ, exported the 3 channels as 3 B&W images, then used the ImageJ analyze feature to draw intensity profiles for the RGB channels of the DP3M standard (left), and the D5000 Standard spectra (right)


Top is Red, middle Green, and bottom blue

The Green region of these spectra run from about 600 (pixels) to 800.

If we look at the Bayer sensor spectra (right side), over 600-800 pixels the Red-Blue ratio is almost constant, So the shades of green in that region will be pretty much unchanged RGB values.

On the other hand, in the Foveon data (left), from 600 to 700 pixels, the red intensity is falling rapidly while the blue is basically zero. And from 700 to 800 pixels, the red is basically zero while the blue is rising rapidly.

So while the red and blue contributions to the recorded color are minor, the red-blue ratio is changing strongly, and this provides the subtle variation in the greens that the Foveon sensor produces.

One has to wonder why the Bayer sensor manufacturers don't use color filters that provide a more gradual cut-off. Years ago, the added computation might have been too much, but I wouldn't think so now.

I don't show the iPhone color responses here, but they have even sharper cut-offs. (And therefore even more plastic-looking colors.)

David

 

[Stillton]

by the way, these are spectra of monitors/displays I have at home:

 

[jande9]

Violet is a hard colour to capture because it is a shorter wavelength than all three of the sensors in a camera or cones in our eyes are sensitive to.

In our eye the red sensitive cone dips down to nothing towards the green, and then bumps up again past the blue, so violet is sensed by the blue and the red cones.

It looks like the foveon uses the blue and the green layers to calculate violet.

 

[xpatUSA]

Violet is a hard colour to capture because it is a shorter wavelength than all three of the sensors in a camera or cones in our eyes are sensitive to.

Yes, Violet light has a wavelength between approximately 380 and 435 nanometers.

https://en.wikipedia.org/wiki/Violet_(color)#Optics

Say 410nm.



So maybe 410nm is about x,y = 0.18, 0.02 just barely within CIE 1931 RGB.

In our eye the red sensitive cone dips down to nothing towards the green, and then bumps up again past the blue, so violet is sensed by the blue and the red cones.

Yes, the so-called color-matching functions:

It looks like the foveon uses the blue and the green layers to calculate violet.

Now the early Foveon response:


At 410nm, a bit of red, a bit of green and a lot of blue

and finally an early raw to XYZ conversion matrix:

All we have to do now is to apply Fig.7 to the matrix and see what XYZ comes out ...

... anyone?

 

[Stillton]

Violet is a hard colour to capture because it is a shorter wavelength than all three of the sensors in a camera or cones in our eyes are sensitive to.

In our eye the red sensitive cone dips down to nothing towards the green, and then bumps up again past the blue, so violet is sensed by the blue and the red cones.

It looks like the foveon uses the blue and the green layers to calculate violet.

Given that blue and green are two "top" layers, and red is the lowest layer, it makes sense. Red wont be capturing any violet anyway since all that 380-400nm goodness will be absorbed by first two layers anyway.

 

[xpatUSA]

Violet is a hard colour to capture because it is a shorter wavelength than all three of the sensors in a camera or cones in our eyes are sensitive to.

In our eye the red sensitive cone dips down to nothing towards the green, and then bumps up again past the blue, so violet is sensed by the blue and the red cones.

It looks like the foveon uses the blue and the green layers to calculate violet.

Given that blue and green are two "top" layers, and red is the lowest layer, it makes sense. Red wont be capturing any violet anyway since all that 380-400nm goodness will be absorbed by first two layers anyway.

Apparently this model of Foveon (F7) does capture some Violet:

 

[saltydogstudios]

Amazing work, thanks for sharing.

Nice to confirm that Nikon's red sensels pick up quite a bit of blue spectrum somewhat mimicking the human eye. Looks like Foveon does that too, but much closer to the response curve of the human eye.

I'm surprised how much the blue sensels in both pick up in the red range - and how much stronger the Foveon's response is than Nikon's. I don't really see this on human eye sensitivity charts, but it makes sense re: our perception of violet.

As for why only Foveon picks up violet - I'm not surprised, but interesting to see in the data how the blue sensel response basically cuts off at higher frequencies of light - so the red sensels pick it up but the blue sensels cut off sharply right around that range, leaving only the red sensels to pick up that range. It seems like a deliberate choice by Nikon based on how sharply it cuts off.

IMO Nikon's response curves lead to "pleasing" skin tones, which I describe as relatively little variation around the main colors of skin - yellowish from melanin and reddish/blueish, from where the blood vessels are closer to the surface. Leica is the worst at this, I've seen numerous photos from Leica where skin looks overly blotchy. Canon and Sony strike a nice balance and Fuji is somewhat more on the Nikon side of things.

I hope whenever a full frame Foveon comes out, Sigma touts these features rather than just focusing on megapixel equivalence or whatever.

--
"no one should have a camera that can't play Candy Crush Saga."
https://www.instagram.com/sodiumstudio/
Camera JPG Portrait Shootout http://sodium.nyc/blog/2020/05/camera-jpg-portrait

 

[jande9]

Part 1 - Foveon colors through a prism of a spectroscope DP3M: Sigma Camera Talk Forum: Digital Photography Review (dpreview.com)
After much f**ing around, I finally realized that I CAN take images of spectra from my spectroscope directly - by putting its ocular against the lens of a camera. That way the images come out a bit clearer and more contrast-y.
Anyway, here are the results:





Note1: vertical lines are not defects/artifacts but rather absorption lines of various elements in the atmosphere. These were super handy as I used them to align the results.
Note2: iPhone8 image has excellent contrast, because its lens was smaller than the ocular allowing for a nice optical seal.
This was not the case with my D5000 or DP3M, so some light was still getting into the lens and reducing contrast.

Its hard to figure out what is happening. CFA cameras do a great job capturing all the subtle variations in colour, but here it seems the monochromatic light, the filters, and the algorithms aren't getting along. Weird.

Jan

 

[saltydogstudios]

Some additional discussion in the Nikon forum.

https://www.dpreview.com/forums/thread/4733919

 

[Stillton]

Some additional discussion in the Nikon forum.

https://www.dpreview.com/forums/thread/4733919

Excellent! My plan to take over the world through measuring spectra seems to be working.

 

[xpatUSA]

Some additional discussion in the Nikon forum.

https://www.dpreview.com/forums/thread/4733919

Excellent! My plan to take over the world through measuring spectra seems to be working.

Me too ...


3500K LED floodlamp through a home-made spectroscope ...

:-D