CCD and CMOS colour

[Erik Kaffehr]

I sometimes see claims that the old CCD sensors gave images with "better" colour quality than CMOS sensors.

If so, why ? I can't think of any obvious reason. Perhaps there was a change in the choice of dyes for the CFA that happened to coincide with the introduction of CMOS ?

Don Cox

Hi,

I would strongly suggest it's a myth.

There is a great probability that color profiles, that is the math describing the conversion of sensor RGB to some well defined color space like XYZ or LAB play a major role that sensor designs.

There is a related myth that 'strong' CFA (color filter array) are beneficial for color, but that is probably not correct.

In the early digital era, most sensors were CCD. Some vendors used CCD sensors from Kodak and those sensors used CFA designs from Kodak,

Phase One later switched to DALSA and DALSA probably had different CFA designs from Kodak.

There were a couple of articles on the issue at On Landscape. Tim Parkin and Joe Cornish suggested that Phase One P45+ and Hasselblad HxD39 had bad reproduction of vegetable greens, not being to be able to separate chlorophyll A and chlorophyll B.

Tim Parkin suggested that it was consistent with the SMI (Sensitivity Metemerism Index) that was very low, 72. later Jim Cornish switched to a DALSA based system and those problems went away. That camera had an SMI of 80.

It has been suggested by some quite respected experts in the field, that the Sony Alpha 900 had the best color rendition in that era, with SMI at 87.

A few years ago, Phase One introduced a new MFD back with what they called new technology. At the same time they have made a presentation describing the new sensor, that was mostly fake, but that initiated some decent studies:

Loading…

www.strollswithmydog.com

Somewhat related, I did a small test at that time:

Loading…

forum.luminous-landscape.com

The question was weather viewers would be able to identify and MFD CFD back with CMOS on 24x36 mm.



The systems were

A) P45+
B) Sony A7rII
C) Sony Alpha 900

So, most viewers identified the A900, which was Sony's first 24x36 CMOS camera as being P45+ (that CCD based MFD.

In this test, all cameras were compared using individual profiles created with LumaRiver Profile Designer





Personally, I would suggest that color profiles can have a lot of tweaks for color and those tweaks dominate over CFA design, while CCD and CMOS does not affect color at all.

Most sensors have IR filters over the sensor, also called hot mirror. The hot mirror may affect color rendition.

The 800 pound gorilla is white balance. Even small changes of WB can have huge effect on rendition.

Best regards

Erik

--
Erik Kaffehr
Website: http://echophoto.dnsalias.net
Magic uses to disappear in controlled experiments…
Gallery: http://echophoto.smugmug.com
Articles: http://echophoto.dnsalias.net/ekr/index.php/photoarticles

 

[Iliah Borg]

It seems manufacturers have chosen "weaker" CFAs due to marketing pressure to be better with low light sensitivity. By weaker I think we all understand CFA dyes that let through more light away from the dye's peak color.

"Weaker" doesn't mean much from the point of view of colour accuracy. It doesn't describe well enough the real issue which is the error while converting from raw to XYZ.

So filters that are more transmissive overall and less selective.

Years ago I had an EE professor who was part of the NTSC color TV committee and had been in on the foundational work. After much begging from us (which was encouraged by the other faculty) we got him to give us a couple of extra evening lectures on NTSC. He described a lot. One thing that stuck with me was his descriptions of the human testing on color perception and how hard they worked to get the color accuracy as good as it could possibly be, and to cover as large of a range of colors as they could. It also involved picking the ideal "standard' phosphors for the color TVs (with cathode ray tubes of course). With the standard phosphors, colors could be really good, and early TVs had them.

Trouble was the standard phosphors couldn't get as bright as some others. When a consumer chose a TV in a store, particularly a store with bright lights, with side by side comparisons, almost always a consumer would skip over the more color accurate TV and would chose the less color accurate TV because it was brighter.

Matched for brightness, a consumer would pick the more color accurate TV every single time. But that was not how the TVs were used in a living room. Our professor held that up to us as an example of engineers working really hard on the wrong problem, or at least not understanding the consumer perspective. Although the committee had included living room characterizations and expected viewing habits, based on movie theaters, as part of the investigations, which he also described to us. But it was misleading. Just because people happily sat in a darkened theater to watch a movie didn't mean they wanted to sit in a darkened living room to watch TV.

The color orange was particularly problematic with the brighter phosphors. One of the things I think helped persuade our professor to give us the extra lectures was us asking about orange. "Why was it the dingy orange we saw on college football uniforms in real life didn't look anything like the bright orange we saw on TV." He kind of snapped at us that the uniform colors were formulated to look good on TV, not in real life. After that, I think he felt obligated to give us an extensive explanation, and that involved mathematical equations and CIE diagrams.

 

[Iliah Borg]

Modern CMOS sensors generally have weaker CFA's and so must use stronger RAW transforms in order to recreate the color faithfully in the scene. Still many find colors from CMOS cameras to be somewhat muted compared to their CCD brethren. When one raises vibrancy or saturation to bring a CMOS image to CCD saturation levels it often means overcooking it’s strong colors as the weak colors are brought to parity.

Just want to be sure no one thinks this is a fundamental difference. This tradeoff is market-driven. Modern CMOS image sensors tend to have smaller pixels (to increase resolution and reduce optics weight, volume and cost) and thus. less light gathering capability per pixel. A "weaker" CFA is used to partially compensate this.

Do the same with a CCD and you will also get "weak colors." Except of course it will take a longer time to readout a high resolution CCD image, not to mention other drawbacks.

Less light gathering capability means more noise. But "weaker" Kodak CMY CFAs over CCDs were abandoned because "strong" colour transforms were bringing the noise back. Same reason for abandoning CMYG, RGBE, RGBW CFAs.

"Weaker" and "stronger" are a matter of colorimetry, not just density. Output saturation is the matter of profile and noise. Colour accuracy is mostly not a matter at all; as Kodak used to say, we give you the colours that you want, not the accurate colours. Consumer market doesn't demand accurate colour, only the likable colour. The issue sometimes is metameric errors and separation of close hues.

 

[Eric Fossum]

It seems manufacturers have chosen "weaker" CFAs due to marketing pressure to be better with low light sensitivity. By weaker I think we all understand CFA dyes that let through more light away from the dye's peak color.

"Weaker" doesn't mean much from the point of view of colour accuracy. It doesn't describe well enough the real issue which is the error while converting from raw to XYZ.

So filters that are more transmissive overall and less selective.

I think this last statement: "more transmissive overall and less selective" means a lot. The extreme case would be no selectivity (i.e. no filter) in which I case I do believe color accuracy would suffer considerably, and which I judge to be the more fundamental problem. Conversion issues follow from that.

 

[Iliah Borg]

It seems manufacturers have chosen "weaker" CFAs due to marketing pressure to be better with low light sensitivity. By weaker I think we all understand CFA dyes that let through more light away from the dye's peak color.

"Weaker" doesn't mean much from the point of view of colour accuracy. It doesn't describe well enough the real issue which is the error while converting from raw to XYZ.

So filters that are more transmissive overall and less selective.

I think this last statement: "more transmissive overall and less selective" means a lot.

Yes, if one doesn't imply causation.

Selectivity when it comes to differentiating close hues is not in the colorimetry everyday vocabulary, so without a definition (which will necessary boil down to colorimetric error) it doesn't mean much

The extreme case would be no selectivity (i.e. no filter) in which I case I do believe color accuracy would suffer considerably, and which I judge to be the more fundamental problem. Conversion issues follow from that.

--
http://www.libraw.org/

 

[OpticsEngineer]

You are thinking farther down the processing chain than I was with the words weak and strong as they are used in colorimetry with regard to hue saturation.

In optical engineering, we speak of weak and strong filters in an entirely different way, without thinking about colorimetry. Which again is different from how we speak of weak and strong filters in digital signal processing. Which is different that how we speak of weak and strong filters with water processing. That was why I started my first post with a definition, actually a restatement, of how the word "weak" filter was already being used by others in this thread.

 

[Iliah Borg]

You are thinking farther down the processing chain than I was with the words weak and strong as they are used in colorimetry with regard to hue saturation.

In optical engineering, we speak of weak and strong filters in an entirely different way, without thinking about colorimetry.

Yes. But that's where we don't need to assume "weaker filters ergo colour". Colour is the result of things that happen further down

Which again is different from how we speak of weak and strong filters in digital signal processing. Which is different that how we speak of weak and strong filters with water processing. That was why I started my first post with a definition, actually a restatement, of how the word "weak" filter was already being used by others in this thread.

Yes, but the problem is not in the word itself, it's in cause-effect inference

 

[JimKasson]

It seems manufacturers have chosen "weaker" CFAs due to marketing pressure to be better with low light sensitivity. By weaker I think we all understand CFA dyes that let through more light away from the dye's peak color.

"Weaker" doesn't mean much from the point of view of colour accuracy. It doesn't describe well enough the real issue which is the error while converting from raw to XYZ.

So filters that are more transmissive overall and less selective.

I think this last statement: "more transmissive overall and less selective" means a lot. The extreme case would be no selectivity (i.e. no filter) in which I case I do believe color accuracy would suffer considerably, and which I judge to be the more fundamental problem. Conversion issues follow from that.

The other extreme case is three filters, each with 1 nm-wide passbands. That produces inaccurate color, too.

 

[D Cox]

It seems manufacturers have chosen "weaker" CFAs due to marketing pressure to be better with low light sensitivity. By weaker I think we all understand CFA dyes that let through more light away from the dye's peak color.

"Weaker" doesn't mean much from the point of view of colour accuracy. It doesn't describe well enough the real issue which is the error while converting from raw to XYZ.

So filters that are more transmissive overall and less selective.

I think this last statement: "more transmissive overall and less selective" means a lot. The extreme case would be no selectivity (i.e. no filter) in which I case I do believe color accuracy would suffer considerably, and which I judge to be the more fundamental problem. Conversion issues follow from that.

The other extreme case is three filters, each with 1 nm-wide passbands. That produces inaccurate color, too.

This is what I call "colour bunching" -- the various hues (or dominant wavelengths) are shifted in the output image toward the dominant wavelengths of the filters.

You can see this in photos of spectra taken with some cameras.

Don Cox

 

[Iliah Borg]

It seems manufacturers have chosen "weaker" CFAs due to marketing pressure to be better with low light sensitivity. By weaker I think we all understand CFA dyes that let through more light away from the dye's peak color.

"Weaker" doesn't mean much from the point of view of colour accuracy. It doesn't describe well enough the real issue which is the error while converting from raw to XYZ.

So filters that are more transmissive overall and less selective.

I think this last statement: "more transmissive overall and less selective" means a lot. The extreme case would be no selectivity (i.e. no filter) in which I case I do believe color accuracy would suffer considerably, and which I judge to be the more fundamental problem. Conversion issues follow from that.

The other extreme case is three filters, each with 1 nm-wide passbands. That produces inaccurate color, too.

That's what I was thinking...

 

[hjulenissen]

The other extreme case is three filters, each with 1 nm-wide passbands. That produces inaccurate color, too.

If there is a "preferred" filter spectral response (the most accurate model of average human beings?), then any deviation from this seems like (in itself) a drawback producing inaccurate color. I believe that:

1. Practical color filter design does not allow arbitrary responses, so one source of undesirable filter response would be the "noise" in settling for physically realilsable filters.

2. When designing filters, spectral sensitivity is only one requirement. Passing through many photons so as to keep luminance noise in check is another. Longevity, thickness, cost, may be other factors.

3. Further, digital color correction consists of both making colors "correct", but also in trade-offs of visible noise and banding.

Really grasping what different CFA responses + different strategies for "color correction" does for the end-to-end response i hard for many photography practitioners and technically minded people as well. What is the range of different filters that are "sort of Luther-Ives compatible", and why would a digital color correction sway away from "perfect" color correction?

-h

 

[JimKasson]

The other extreme case is three filters, each with 1 nm-wide passbands. That produces inaccurate color, too.

If there is a "preferred" filter spectral response (the most accurate model of average human beings?), then any deviation from this seems like (in itself) a drawback producing inaccurate color. I believe that:

Luckily, it doesn’t have to do that, it just has to be a 3x3 matrix multiply away from that.

1. Practical color filter design does not allow arbitrary responses, so one source of undesirable filter response would be the "noise" in settling for physically realilsable filters.

2. When designing filters, spectral sensitivity is only one requirement. Passing through many photons so as to keep luminance noise in check is another. Longevity, thickness, cost, may be other factors.

3. Further, digital color correction consists of both making colors "correct", but also in trade-offs of visible noise and banding.

Really grasping what different CFA responses + different strategies for "color correction" does for the end-to-end response i hard for many photography practitioners and technically minded people as well. What is the range of different filters that are "sort of Luther-Ives compatible",

It is infinite, as is the range of filters that is precisely LI compatible.

and why would a digital color correction sway away from "perfect" color correction?

-h

Market forces. With four dyes in the CFA you can do a lot better than with three, but that approach has not gotten any traction to speak of.

 

[Iliah Borg]

The other extreme case is three filters, each with 1 nm-wide passbands. That produces inaccurate color, too.

If there is a "preferred" filter spectral response (the most accurate model of average human beings?), then any deviation from this seems like (in itself) a drawback producing inaccurate color. I believe that:

1. Practical color filter design does not allow arbitrary responses, so one source of undesirable filter response would be the "noise" in settling for physically realilsable filters.

2. When designing filters, spectral sensitivity is only one requirement. Passing through many photons so as to keep luminance noise in check is another. Longevity, thickness, cost, may be other factors.

3. Further, digital color correction consists of both making colors "correct", but also in trade-offs of visible noise and banding.

Really grasping what different CFA responses + different strategies for "color correction" does for the end-to-end response i hard for many photography practitioners and technically minded people as well. What is the range of different filters that are "sort of Luther-Ives compatible", and why would a digital color correction sway away from "perfect" color correction?

-h

To add one more consideration to what Jim said:

Neither displays no printers satisfy LI condition. Minimizing metameric error towards typical output devices and viewing conditions is an important thing.

 

[John Sheehy]

The old CFA's were clearly built to prioritize color fidelity at base ISO, the new CFA's seem to be more prioritized for high resolution sensors and high-ISO performance.

There is also the fact that these claimed "better" colors happened across a range of white balance needs that are more varied than the difference in spectral response between CCD and CMOS monochrome sensors.

 

[John Sheehy]

It seems manufacturers have chosen "weaker" CFAs due to marketing pressure to be better with low light sensitivity. By weaker I think we all understand CFA dyes that let through more light away from the dye's peak color.

"Weaker" doesn't mean much from the point of view of colour accuracy. It doesn't describe well enough the real issue which is the error while converting from raw to XYZ.

Neutral density certainly qualifies as "weaker" but has no effect on color per se; just on noise.

 

[Iliah Borg]

For those interested, here is a link to an older post you made with spectral curves of an earlier and later CFAs

Re: Still awaiting your response...: Open Talk Forum: Digital Photography Review (dpreview.com)

To interpret those graphs one needs to construct transforms to colour and compare the output. Without this, even if the graphs are correct, they can't be interpreted.

 

[Iliah Borg]

It seems manufacturers have chosen "weaker" CFAs due to marketing pressure to be better with low light sensitivity. By weaker I think we all understand CFA dyes that let through more light away from the dye's peak color.

"Weaker" doesn't mean much from the point of view of colour accuracy. It doesn't describe well enough the real issue which is the error while converting from raw to XYZ.

Neutral density certainly qualifies as "weaker" but has no effect on color per se; just on noise.

I'm just trying to say that "weaker" or "stronger" (more dense) are the terms that don't predict colour separation well.

 

[JimKasson]

It seems manufacturers have chosen "weaker" CFAs due to marketing pressure to be better with low light sensitivity. By weaker I think we all understand CFA dyes that let through more light away from the dye's peak color.

"Weaker" doesn't mean much from the point of view of colour accuracy. It doesn't describe well enough the real issue which is the error while converting from raw to XYZ.

Neutral density certainly qualifies as "weaker" but has no effect on color per se; just on noise.

I'm just trying to say that "weaker" or "stronger" (more dense) are the terms that don't predict colour separation well.

True. Except in the limit, where they are both bad inaccurate.

--
https://blog.kasson.com

 

[Iliah Borg]

It seems manufacturers have chosen "weaker" CFAs due to marketing pressure to be better with low light sensitivity. By weaker I think we all understand CFA dyes that let through more light away from the dye's peak color.

"Weaker" doesn't mean much from the point of view of colour accuracy. It doesn't describe well enough the real issue which is the error while converting from raw to XYZ.

Neutral density certainly qualifies as "weaker" but has no effect on color per se; just on noise.

I'm just trying to say that "weaker" or "stronger" (more dense) are the terms that don't predict colour separation well.

True. Except in the limit, where they are both bad inaccurate.

But artificial intellect can do miracles, expanding the limits. Soon we will have accurate colour from monochrome, so they say.

 

[J A C S]

OpticsEngineer, post: 65422247, member: 1261637"]
It seems manufacturers have chosen "weaker" CFAs due to marketing pressure to be better with low light sensitivity. By weaker I think we all understand CFA dyes that let through more light away from the dye's peak color.

"Weaker" doesn't mean much from the point of view of colour accuracy. It doesn't describe well enough the real issue which is the error while converting from raw to XYZ.

Neutral density certainly qualifies as "weaker" but has no effect on color per se; just on noise.

I'm just trying to say that "weaker" or "stronger" (more dense) are the terms that don't predict colour separation well.

True. Except in the limit, where they are both bad inaccurate.
[/QUOTE]
I asked fPrime several times to define "weaker" CFA, and he never did...

BTW, Fuji's filters come in several thicknesses according to this page. Assuming that they are made of the same material, just the thickness changes, the transmission is raised to some power; for example each 1.1 μm curve has to be raised to power 1.1/0.7 to get the corresponding 0.7 μm one. To make things simple, assume filter B twice as thick as filter A. Then when A transmits 90%, B would transmit 81%. When A transmits 10%, B would transmit 1% (!). This makes B "stronger" - more concentrated near the peak(s).