Visual depiction of light loss in Bayer CFA

Started Sep 30, 2010 | Discussions
Joofa Senior Member • Posts: 2,783
Visual depiction of light loss in Bayer CFA
1

Motivated by some discussion on this forum regarding light losses in a Bayer CFA I did a visual experiment by taking two exact same cameras with the same lens (F/2.8), exposure (300 msec), gain (ISO), readout speed, etc., with a 16M sensor with and without a Bayer CFA. No anti-aliasing filter (AA) was applied to any sensor. The result is below:

The image on the left is the Bayer CFA color image presented as raw data (i.e, no conversion first to color and then to grayscale was done), where as the image on the right is from the sensor without the Bayer CFA.

Enjoy,

Joofa

-- hide signature --

Dj Joofa

godfrog Senior Member • Posts: 1,463
Re: Visual depiction of light loss in Bayer CFA

Interesting, what type of light source are you using, and is there any UV or IR filters in use?

-- hide signature --
OP Joofa Senior Member • Posts: 2,783
Re: Visual depiction of light loss in Bayer CFA

godfrog wrote:

Interesting, what type of light source are you using, and is there any UV or IR filters in use?

Ordinary tube lights. No UV / IR / AA filters were used.

Joofa

-- hide signature --

Dj Joofa

Daniel Browning Senior Member • Posts: 1,058
Wow

That is quite a big difference. Thanks for posting.
--
Daniel

Photato
Photato Senior Member • Posts: 2,811
How did you get this B&W outfit ?

Care to elaborate how did you get that camera without the color filters ?
What Brand/Model are these ?

Comparing Histograms the One on the right is from +1 to +1.8 Brighter.

Now, can you make two WELL EXPOSED shots to compare Dynamic Range, with subject being a a Wide Range in light values ?
I bet the RGB color camera wins there.

 Photato's gear list:Photato's gear list
Panasonic LX100 Canon EOS M Canon EOS M6 Canon EOS M6 II Canon EF-M 11-22mm f/4-5.6 IS STM +13 more
Lee Jay Forum Pro • Posts: 56,673
Was gamma appled to these images?

These look a little weird to me unless they are close to linear data.

-- hide signature --

Lee Jay
(see profile for equipment)

 Lee Jay's gear list:Lee Jay's gear list
Canon PowerShot SX50 HS Canon EOS 7D Mark II Canon EF 70-200mm F2.8L IS II USM Canon Extender EF 2x III Canon EF 8-15mm f/4L Fisheye USM +10 more
Simon97
Simon97 Veteran Member • Posts: 3,214
Re: Visual depiction of light loss in Bayer CFA

Could you post a 100% crop of a focused part of the image? Just curious of what it looks like.
Thanks

 Simon97's gear list:Simon97's gear list
Canon PowerShot ELPH 360 HS Panasonic Lumix DMC-GX85
Barrie Davis
Barrie Davis Forum Pro • Posts: 21,460
Re: Visual depiction of light loss in Bayer CFA

Joofa wrote:

The image on the left is the Bayer CFA color image presented as raw data (i.e, no conversion first to color and then to grayscale was done), where as the image on the right is from the sensor without the Bayer CFA.

Well, one would expect approximately 2/3rds of the light to be absorbed by an RGB Bayer filter array, so a gain of around 150% would be anticipated if it was taken off.

Yeah... looks about right.

Makes me wonder why CMY Bayer arrays have not been persued my assiduously. In theory they would double the sensitivity of RGB types.
--
Regards,
Baz

"Ahh... But the thing is, they were not just ORDINARY time travellers!"

 Barrie Davis's gear list:Barrie Davis's gear list
Konica Minolta DiMAGE A2
Lee Jay Forum Pro • Posts: 56,673
Re: Visual depiction of light loss in Bayer CFA

Barrie Davis wrote:

Joofa wrote:

The image on the left is the Bayer CFA color image presented as raw data (i.e, no conversion first to color and then to grayscale was done), where as the image on the right is from the sensor without the Bayer CFA.

Well, one would expect approximately 2/3rds of the light to be absorbed by an RGB Bayer filter array, ...

The spectral response of these dyes are quite wide and overlap quite a lot. That's why the generalized assumption is more like half than 2/3.

-- hide signature --

Lee Jay
(see profile for equipment)

 Lee Jay's gear list:Lee Jay's gear list
Canon PowerShot SX50 HS Canon EOS 7D Mark II Canon EF 70-200mm F2.8L IS II USM Canon Extender EF 2x III Canon EF 8-15mm f/4L Fisheye USM +10 more
OP Joofa Senior Member • Posts: 2,783
Combined response

Daniel Browning wrote:

That is quite a big difference. Thanks for posting.

You are welcome. Hi, the actual images are quite clean, as opposed to the smaller sized images I displayed in my original message, so I redid the images in ImageJ. Please reload in your browser. But I'm also providing it again below. Please scroll to the right to see the histogram and other statistics as shown by ImageJ.

Photato wrote:

Care to elaborate how did you get that camera without the color filters ?
What Brand/Model are these ?

Well, if you want to see the camera, then an image is below. The only difference is that when I took the picture of the camera below I had a C-mount on it as seen by that tiny lens. So for this experiment I removed the front plate and put on an F-mount and used a Nikon lens to go with the 16 MP, 24mmx36mm sensor. I have both Bayer CFA and a monochrome (without CFA) versions of the sensor, so switching them out is easy.

Comparing Histograms the One on the right is from +1 to +1.8 Brighter.

I have the histogram statistics in the image embedded above.

ljfinger wrote:

These look a little weird to me unless they are close to linear data.

Yes, they are linear, raw data presented in both, without any gamma.

Joofa

-- hide signature --

Dj Joofa

photoSmart42 Senior Member • Posts: 1,380
very intresting!

Thanks for posting this!

Lee Jay Forum Pro • Posts: 56,673
Re: Combined response

Joofa wrote:

Yes, they are linear, raw data presented in both, without any gamma.

Ah...that explains a lot. Is there any chance you could apply a plain gamma=2 tone curve to them just so they look a little more like what we're used to?

-- hide signature --

Lee Jay
(see profile for equipment)

 Lee Jay's gear list:Lee Jay's gear list
Canon PowerShot SX50 HS Canon EOS 7D Mark II Canon EF 70-200mm F2.8L IS II USM Canon Extender EF 2x III Canon EF 8-15mm f/4L Fisheye USM +10 more
hjulenissen Senior Member • Posts: 2,315
Re: Visual depiction of light loss in Bayer CFA

godfrog wrote:

Interesting, what type of light source are you using, and is there any UV or IR filters in use?

Do you think that as long as a "human luminance sensitivity" wavelength filtering was to be applied (no UV, no IR, most sensitive to green) and a "common" light source (such as daylight), the difference would be much smaller?

-h

Photato
Photato Senior Member • Posts: 2,811
Re: Combined response

Well, thanks for the prompt response and posting the picture of your gear.
So it looks more like +2EV in extra sensitivity for the B&W sensor.

What I still doubt is the Dynamic Range response of the B&W sensor.

How about making a comparison like that. I know the Color Sensor will have more of it, but not sure how big is that gap.

Thanks for this interesting thread.

BTW I have that same Laptop. Lenovo Thinkpad T61P with the Dockstation, great Computer, awful Screen. Haha !

 Photato's gear list:Photato's gear list
Panasonic LX100 Canon EOS M Canon EOS M6 Canon EOS M6 II Canon EF-M 11-22mm f/4-5.6 IS STM +13 more
Barrie Davis
Barrie Davis Forum Pro • Posts: 21,460
Re: Visual depiction of light loss in Bayer CFA

ljfinger wrote:

Barrie Davis wrote:

Well, one would expect approximately 2/3rds of the light to be absorbed by an RGB Bayer filter array, ...

The spectral response of these dyes are quite wide and overlap quite a lot. That's why the generalized assumption is more like half than 2/3.

Yes. I wouldn't argue with that. And estimates of the brightness gain on the sample images are likely just as appoximate.
--
Regards,
Baz

"Ahh... But the thing is, they were not just ORDINARY time travellers!"

 Barrie Davis's gear list:Barrie Davis's gear list
Konica Minolta DiMAGE A2
Lee Jay Forum Pro • Posts: 56,673
Linear data

Photato wrote:

Well, thanks for the prompt response and posting the picture of your gear.
So it looks more like +2EV in extra sensitivity for the B&W sensor.

If you're doing this estimate by eye, or with tools normally used for gamma-corrected data (which is most of the usual suspects), I would caution against that conclusion because we're looking at linear data.

-- hide signature --

Lee Jay
(see profile for equipment)

 Lee Jay's gear list:Lee Jay's gear list
Canon PowerShot SX50 HS Canon EOS 7D Mark II Canon EF 70-200mm F2.8L IS II USM Canon Extender EF 2x III Canon EF 8-15mm f/4L Fisheye USM +10 more
panos_m Senior Member • Posts: 1,621
Re: Linear data

ljfinger wrote:

Photato wrote:

Well, thanks for the prompt response and posting the picture of your gear.
So it looks more like +2EV in extra sensitivity for the B&W sensor.

If you're doing this estimate by eye, or with tools normally used for gamma-corrected data (which is most of the usual suspects), I would caution against that conclusion because we're looking at linear data.

Hard to tell. Checking linear 8bit values I would say something around 1,5 stops difference. Very interesting.
--
Panagiotis

Lee Jay Forum Pro • Posts: 56,673
Re: Linear data

panos_m wrote:

ljfinger wrote:

Photato wrote:

Well, thanks for the prompt response and posting the picture of your gear.
So it looks more like +2EV in extra sensitivity for the B&W sensor.

If you're doing this estimate by eye, or with tools normally used for gamma-corrected data (which is most of the usual suspects), I would caution against that conclusion because we're looking at linear data.

Hard to tell. Checking linear 8bit values I would say something around 1,5 stops difference. Very interesting.

That's about what I get using linear as well. That's surprising to me unless the Bayer dyes in the Bayer sample are somewhat non-typical of consumer cameras. Perhaps Barrie's 2/3 of total light estimate above was correct after all!

-- hide signature --

Lee Jay
(see profile for equipment)

 Lee Jay's gear list:Lee Jay's gear list
Canon PowerShot SX50 HS Canon EOS 7D Mark II Canon EF 70-200mm F2.8L IS II USM Canon Extender EF 2x III Canon EF 8-15mm f/4L Fisheye USM +10 more
Joseph S Wisniewski Forum Pro • Posts: 34,385
Interesting. Did you have a photopic filter?

Joofa wrote:

Daniel Browning wrote:

That is quite a big difference. Thanks for posting.

You are welcome. Hi, the actual images are quite clean, as opposed to the smaller sized images I displayed in my original message, so I redid the images in ImageJ. Please reload in your browser. But I'm also providing it again below. Please scroll to the right to see the histogram and other statistics as shown by ImageJ.

http://www.djjoofa.com/data/images/bw_col_both.jpg

Indeed. Thank you very much.

It's a comparison I can't run, myself. My monochrome sensors don't have microlenses, while my color one of the same type does.

I liked the aliased images. They showed the pattern of the Bayer CFA

Seriously, I was able to read something from that. I assume the thing that looks like a whiteboard really was white. You're showing the green pixels as being significantly brighter than the red or blue (not making any guesses as to whether the darkest ones are the red or the blue, however).

So, if that's linear, and the mean of the monochrome image is 3.3 that of the Bayer image (which is hard to say, as the monochrome looks blown) then I'm guessing something like the monochrome being about 2x higher than the green and maybe 4-5x higher than the red or blue.

That's pretty reasonable, if you didn't have a photopic filter (or even some sort of blue-green "red and IR killer" like a Schott BG-34) on the monochrome sensor.

I'm also guessing fluorescent light, because tungsten or xenon strobe would have put the red channel up closer to the green...

Photato wrote:

Care to elaborate how did you get that camera without the color filters ?
What Brand/Model are these ?

Well, if you want to see the camera, then an image is below. The only difference is that when I took the picture of the camera below I had a C-mount on it as seen by that tiny lens. So for this experiment I removed the front plate and put on an F-mount and used a Nikon lens to go with the 16 MP, 24mmx36mm sensor. I have both Bayer CFA and a monochrome (without CFA) versions of the sensor, so switching them out is easy.

Can I ask what you were using, at 16mp in 24x36?

Comparing Histograms the One on the right is from +1 to +1.8 Brighter.

I have the histogram statistics in the image embedded above.

-- hide signature --

Rahon Klavanian 1912-2008.

Armenian genocide survivor, amazing cook, scrabble master, and loving grandmother. You will be missed.

Ciao! Joseph

http://www.swissarmyfork.com

 Joseph S Wisniewski's gear list:Joseph S Wisniewski's gear list
Nikon D3 Nikon D2X Nikon D90 Nikon D100 Nikon Z7 +45 more
OP Joofa Senior Member • Posts: 2,783
Re: Interesting. Did you have a photopic filter?

Joseph S Wisniewski wrote:

Indeed. Thank you very much.

You are most welcome.

I liked the aliased images. They showed the pattern of the Bayer CFA

Yes.

Seriously, I was able to read something from that. I assume the thing that looks like a whiteboard really was white.

Yes, white.

So, if that's linear, and the mean of the monochrome image is 3.3 that of the Bayer image (which is hard to say, as the monochrome looks blown) then I'm guessing something like the monochrome being about 2x higher than the green and maybe 4-5x higher than the red or blue.

I think so. I shall see if I can put up the raw file so that people can download it.

That's pretty reasonable, if you didn't have a photopic filter (or even some sort of blue-green "red and IR killer" like a Schott BG-34) on the monochrome sensor.

I didn't use any filter.

I'm also guessing fluorescent light, because tungsten or xenon strobe would have put the red channel up closer to the green...

Yes, usual tube light. In another image that I acquired in a 10 nm bandwidth in Tungsten-like light some time ago actually shows red software softer (what was I thinking when I wrote software) and less noisy than green:

http://forums.dpreview.com/forums/read.asp?forum=1018&message=36490615

Can I ask what you were using, at 16mp in 24x36?

You mean what I am using this sensor for?

Joofa

-- hide signature --

Dj Joofa

Keyboard shortcuts:
FForum MMy threads