[rechmbrs]

The Color Filter Array on the sensor gets little comment but I think the makeup of CFA is very important. I remember back some time ago that the color filtering on the cells of the sensor vary from camera manufacturer to manufacturer. Say for the sensor in the K 1 and the Nikon D8xx have CFA designed by Pentax and Nikon respectively.
What I don't know is how much of that glass? is from Pentax? Is the effect of different CFA equivalent to the difference in film? Is the CFA difference used in computing color correction? Is this one of the reasons that out of the camera jpg are so different from manufacturer to manufacturer?

RONC

[stevebrot]

What I don't know is how much of that glass? is from Pentax? Is the effect of different CFA equivalent to the difference in film?

It is my understanding that the filter array is assembled into the sensor prior to microlens application during manufacture at Sony. Whether it is provided or even spec'ed by Ricoh is a good question. I would vote not, but who knows what options are exposed to customers by Sony.

Edit: I just some research on this and found that the microlens and color filter array are probably a single bonded unit. The article linked below has some interesting photos of the layers being removed for astrophotography purposes.

https://petapixel.com/2013/08/04/scratching-the-color-filter-array-layer-off...per-bw-photos/

Is this one of the reasons that out of the camera jpg are so different from manufacturer to manufacturer?

That is hard to say since the photosites themselves are also open to tuning (both pre and post capture) and color rendition is also one of the major function of demosaic. I would vote not.

I am not an engineer, but if I were, my design efforts in this area would be to optimize response (max data) within foreseeable spectral regimes and sort the balance matters out post exposure.


Steve

[photoptimist]

It's hugely important, very complicated, and most certainly affects the color rendering by different cameras much the same way different films differ in color rendering:

1. The choice of colorants affects which wavelengths of light are allowed to pass into the R, G, and B pixels. These are mixtures of chemicals selected to absorb certain frequencies of light while transmitting others. The hard part is that no chemical has the perfect absorption spectrum such as the perfect red dye that lets in all red light but absolutely no green or blue light. (There's also the issue of the color response of the chip which depends on the sensor technology and manufacturing processes.)

2. The density, concentration, or thickness of the filter layer affects the purity or saturation of the colors but also affects the base ISO of the sensor. This creates a big trade-off: compare a deep red filter that gets pure colors but needs a lot of light versus a pink filter that needs less light but doesn't get a pure a color response)

3. The hardware and software that amplifies or mixes the different RGB channels to get a high-fidelity, neutral color response (or a "pleasing" color response) in the demosaiced image.

It's very unlikely that Pentax makes the filters themselves for reasons explained by stevebrot. Instead, Pentax probably specifies the ingredients and thickness of the coatings that are mixed and delivered to Sony for controlled deposition on the sensors. Pentax does have greater control over the hardware and software.

[timw4mail]

Hmm...I wonder why there aren't any consumer cameras that are black and white (without the CFA). I know there are photo backs for some medium format film cameras that are a fully black and white sensor...

[stevebrot]

1. The choice of colorants affects which wavelengths of light are allowed to pass into the R, G, and B pixels. These are mixtures of chemicals selected to absorb certain frequencies of light while transmitting others. The hard part is that no chemical has the perfect absorption spectrum such as the perfect red dye that lets in all red light but absolutely no green or blue light. (There's also the issue of the color response of the chip which depends on the sensor technology and manufacturing processes.)

The filters are most likely dichroic and not dye based.


Steve

[stevebrot]

Hmm...I wonder why there aren't any consumer cameras that are black and white (without the CFA)...

Probably because the market for consumer cameras is not interested in B&W photography

The full monochrome backs/cameras from Phase One are very interesting as is the Leica Monochrom. All three, however, are prohibitively expensive. I have wondered, though whether pixel-shift might be leveraged to generate a high quality monochrome image. After all, every pixel has the benefit of full chromatic data (Bayer array is defeated).


Steve

[photoptimist]

The filters are most likely dichroic and not dye based.


Steve

Hmmm.. I very much doubt that. Although dichroic filters are absolutely amazing in terms of allowing high transmission of exactly the color you want and near zero transmission of unwanted colors, they suffer from two serious problems in this application. First, they are extremely reflective. Instead of absorbing the unwanted color, they reflect it back creating significant flare issues. Second, their color response is angle dependent with the band pass wavelength shifting by 1/cos(theta). In the context of ILCs, that means the color response of the sensor would vary from center to corner, be different for different lenses, and also change with aperture. With a large aperture lens, you'd even see color shifts across each bokeh circle.

Dichroic filters are sometimes used in film scanners (including the ones that I designed) but in those cases the flare is no problem (e.g., when the red filter is in place, the cyan light it reflects light can't get through it even if it bounces around in the optical system) and the angle issue is correctable because the scanners have a fixed lens, fixed aperture, and pretty modest ray angles.

[rechmbrs]

Boy has this turned up a wealth of knowledge. Thanks all.

I still would like to know how much control that Pentax has in CFA building process for the K 1. Any one of you spied a set of spectral responses for k1 and d8xx CFA? Why do I want to know? I'm seeing many more images with multiple differing light sources and would like to know how to handle the color balancing best method.

Thanks,
RONC

[stevebrot]

Hmmm.. I very much doubt that. Although dichroic filters are absolutely amazing in terms of allowing high transmission of exactly the color you want and near zero transmission of unwanted colors, they suffer from two serious problems in this application. First, they are extremely reflective. Instead of absorbing the unwanted color, they reflect it back creating significant flare issues. Second, their color response is angle dependent with the band pass wavelength shifting by 1/cos(theta). In the context of ILCs, that means the color response of the sensor would vary from center to corner, be different for different lenses, and also change with aperture. With a large aperture lens, you'd even see color shifts across each bokeh circle.

Dichroic filters are sometimes used in film scanners (including the ones that I designed) but in those cases the flare is no problem (e.g., when the red filter is in place, the cyan light it reflects light can't get through it even if it bounces around in the optical system) and the angle issue is correctable because the scanners have a fixed lens, fixed aperture, and pretty modest ray angles.

Good points. I thought dichroic because of the close integration with the microlens and the obvious thin film nature of the filter.


Steve

[photoptimist]

Boy has this turned up a wealth of knowledge. Thanks all.

I still would like to know how much control that Pentax has in CFA building process for the K 1. Any one of you spied a set of spectral responses for k1 and d8xx CFA? Why do I want to know? I'm seeing many more images with multiple differing light sources and would like to know how to handle the color balancing best method.

Thanks,
RONC

I've never seen them but they aren't hard to measure with a slit light source and diffraction grating or prism.

There's various blogs & thread talking about the issue (see below) and the spectral responses of different cameras but I'v enot seen any specifically for the D8xx or K-1

Re: Ken Rockwell's comments on Sony's color rendition: Sony Alpha / Nex E-mount (APS-C) Talk Forum: Digital Photography Review
A Reference Data Set For Camera*Spectral Sensitivity Estimation | Rank Based Spectral estimation
On chip ADC - other benefits than "just" lower noise/better DR?: Canon EOS-1D / 5D / 6D Talk Forum: Digital Photography Review

[leekil]

The full monochrome backs/cameras from Phase One are very interesting as is the Leica Monochrom. All three, however, are prohibitively expensive. I have wondered, though whether pixel-shift might be leveraged to generate a high quality monochrome image. After all, every pixel has the benefit of full chromatic data (Bayer array is defeated).

The pixelshift should result in high greyscale acuity just as it produces high color acuity -- just convert the color image to b&w the same way one normally would.

[stevebrot]

Why do I want to know? I'm seeing many more images with multiple differing light sources and would like to know how to handle the color balancing best method.

Ohhhhh...deep waters, those are...very deep indeed...


Steve

[stevebrot]

The pixelshift should result in high greyscale acuity just as it produces high color acuity -- just convert the color image to b&w the same way one normally would.

I am more interested in extended tonal fidelity/resolution rather than acuity. Think of it as the monochrome equivalent of so-called color resolution. Think of how B&W film works rather than the spectral response model (RGB retinal cone emulation) used for normal RAW conversion and merge for regular pixel shift. Regular monochrome conversion deals with available data in a fairly blunt manner.


Steve

[photoptimist]

Boy has this turned up a wealth of knowledge. Thanks all.

I still would like to know how much control that Pentax has in CFA building process for the K 1. Any one of you spied a set of spectral responses for k1 and d8xx CFA? Why do I want to know? I'm seeing many more images with multiple differing light sources and would like to know how to handle the color balancing best method.

Thanks,
RONC

As stevebrot said, that's a very deep topic.

There's a very long chain of elements of the system that define and modulate the spectral properties of each and every wavelength of the light:

------- The Scene:
1. the original light source (e.g., the sun, incandescent bulb, LED, CFL, flash, neon lights, etc.)
2. any reflectors (e.g., blue sky, green leaves, a red car next to the bride, the T-shirt of the photographer, a warming reflective disk, etc.)
3. the subject's coloration and reflectivity (e.g., skin, dyes in clothing, pigments in plants, etc.)

------- The Camera:
4. the lens and any filters
5. the Bayer filter
6. the intrinsic detector response of the silicon chip
7. differential electronic or digital amplification of the RGB channels

----- The Post Process:
8. demosaicing (estimates the scene's RGB levels from the measured RAW RGB values)
9. color balance and post processing (saturation, curve manipulation)
10. the output device (monitor, printer)
11. viewing conditions (color of the lights & room around the monitor or print)

Your job as photographer is to get a nice result in step 11 that reflects (ha! pun intended!) the colors of element #3 (the subject) despite all the stuff happening in other steps of the system.

To do that, the first line of action is to control the light source (#1) and reflective items around the subject (#2) with possibly the aid of filter on the lens (#4). A few controlled shots with a color checker can also help calibrate the system (which is what technical photographers do for getting high fidelity images of art and museum pieces).

Of course, many situations don't offer such control (the bride is standing under her favorite green tree next to her red car on the left with blue sky lighting her on the right) which means you're going to be spending a lot of time in step #9 trying to keep the bride from looking like she was splashed by pastel Easter egg colors. (Maybe this is why some photographers shoot black-and-white!)

[leekil]

I am more interested in extended tonal fidelity/resolution rather than acuity. Think of it as the monochrome equivalent of so-called color resolution. Think of how B&W film works rather than the spectral response model (RGB retinal cone emulation) used for normal RAW conversion and merge for regular pixel shift. Regular monochrome conversion deals with available data in a fairly blunt manner.

Steve

I'm not sure what you mean as "color resolution," unless you mean what I meant. I was using "color acuity" to refer to the accurate assignation of color to a single pixel, rather than it "bleeding" into adjacent pixels with the bayer filter limitations -- what pixelshift does. Or are you talking about more bits to represent one pixel?

My intent was that the more accurate color assignation resulting from the pixelshift photo would then enable you to produce a similarly more accurate b&w version of the image. I intentionally left the conversion process undiscussed because there are various more and less accurate conversion options available, but you would still get an improvement in the result in any process, with better input -- the pixelshift image.