Monochrome camera vs converting from color?

[Tim Lookingbill]

Begging the question again, have you ever tried viewing a raw Histogram when attempting to expose anything?

Yes, I have with Fast Raw Viewer.

It didn't tell me anything I could use to mitigate against the unpredictable nature of how the sensor captures detail in intensely colored objects no matter the lighting situation out in the field. There was no set it and forget it button that would tell me if the Raw data got blown or preserved due to the inconsistent nature of sensor response to these kind of vibrant colorful scenes.

[Doug Gray]

Yes, I have with Fast Raw Viewer.

It didn't tell me anything I could use to mitigate against the unpredictable nature of how the sensor captures detail in intensely colored objects no matter the lighting situation out in the field. There was no set it and forget it button that would tell me if the Raw data got blown or preserved due to the inconsistent nature of sensor response to these kind of vibrant colorful scenes.

There is no set and forget it in photography.

Generally, if the RAW sensor values don't clip then they have captured the detail but after that everything depends on your raw converter, camera profile for the converter, and the destination colorspace. Then the main issue is whether you have enough dynamic range.  There are times when clipping the sensor values in RAW are necessary such as when you have light sources or strong specular reflections. For instance taking a picture of a Christmas tree with bright lights. If you don't clip in the RAW the dynamic range required to capture the lights w/o clipping will almost certainly result in insufficient dynamic range.

But the general problem isn't the sensors. Within their range they can capture the entire human visual gamut but monitors and printers have their limits and, when gamut mapping must be done to make things fit, the result can be ugly. That's where the art comes in and the photographer has to select the appropriate camera profile and possibly tweak things in post to produce the most perceptually similar, pleasing, image. There are a lot of choices but no automatic magic.

[Tim Lookingbill]

But the general problem isn't the sensors. Within their range they can capture the entire human visual gamut but monitors and printers have their limits and, when gamut mapping must be done to make things fit, the result can be ugly. That's where the art comes in and the photographer has to select the appropriate camera profile and possibly tweak things in post to produce the most perceptually similar, pleasing, image. There are a lot of choices but no automatic magic.

That's not been my experience shooting and successfully processing over 1000 Raws of varying dynamic ranges, some most hobbyists and professionals wouldn't even attempt to keep.

I've seen no limitations in pulling out detail out of vibrant colored flowers on an sRGB gamut display editing Raw. A custom DNG camera profile didn't make much of a difference but did impede in some wide dynamic range images my pulling out detail where AdobeStandard or ACR 4.4 Adobe canned profiles did a better job. And sometimes the custom DNG profiles worked better. Again there's no consistency or predictability that an optimal Raw exposure can assure I didn't clip the data.

The one thing Fast Raw Viewer did show me is that I've nailed the exposure under 10% error of quite a few Raws of white birds on dark water without establishing an optimal exposure.

[Doug Gray]

Again there's no consistency or predictability that an optimal Raw exposure can assure I didn't clip the data.

Of course.

As long as the RAW file values aren't clipped, and you set the ACR exposure slider to produce the same overall results (push a stop in one, pull in the other), a clipped image will continue to be a clipped image. It has nothing to do with the RAW sensor value gain absent RAW sensor value clipping.

[Jim Kasson]

But the general problem isn't the sensors. Within their range they can capture the entire human visual gamut but monitors and printers have their limits...

Doug, here are some spectral loci with compromise matrices optimized for the Macbeth CC for a few cameras:







In no case that I've found does the recoverable gamut fill the whole horseshoe. Of course, you could create a Procrustean bed with 3D LUTs.

Jim

[Doug Gray]

In no case that I've found does the recoverable gamut fill the whole horseshoe. Of course, you could create a Procrustean bed with 3D LUTs.

Jim

sRGB is a Procrustean Bed!

Realizable sensors and their CFAs can capture the entire human gamut. Just not accurately since they don't meet LI criteria. The "gamut boundary" derived from any given set of color patches is close to meaningless as there are an infinite number of spectra for any one color inside the Macadam's limit. For emissive sources, it's even more expansive as there are infinite spectra sets for all colors inside the human gamut boundary.  Each of these spectra maps as a sort of smear around the specific chromaticity point a true LI condition would produce. And the location of that smear would vary considerably based on the spectra of the colors the matrix was derived from. It wouldn't take too many sets of 18 colored patches and a neutral to derive, with some specified illuminant, a matrix that exceeds the CIExy chart at any arbitrary point. In fact it might be an interesting extended piece of homework for a class of budding color scientists. See who can come up with the smallest set of 24 spectra that have identical CIELAB values as the Colorchecker, which, when used to generate an optimal matrix for a particular CFA/sensor, completely overlap the CIExy chromaticity chart.

[Jim Kasson]

sRGB is a Procrustean Bed!

Realizable sensors and their CFAs can capture the entire human gamut. Just not accurately since they don't meet LI criteria. The "gamut boundary" derived from any given set of color patches is close to meaningless as there are an infinite number of spectra for any one color inside the Macadam's limit. For emissive sources, it's even more expansive as there are infinite spectra sets for all colors inside the human gamut boundary.  Each of these spectra maps as a sort of smear around the specific chromaticity point a true LI condition would produce. And the location of that smear would vary considerably based on the spectra of the colors the matrix was derived from. It wouldn't take too many sets of 18 colored patches and a neutral to derive, with some specified illuminant, a matrix that exceeds the CIExy chart at any arbitrary point. In fact it might be an interesting extended piece of homework for a class of budding color scientists. See who can come up with the smallest set of 24 spectra that have identical CIELAB values as the Colorchecker, which, when used to generate an optimal matrix for a particular CFA/sensor, completely overlap the CIExy chromaticity chart.

Maybe I wasn't clear. The spectral locus for each camera was derived from spectral stimuli 5 nm apart. The gamut boundary was not derived for one set of color patches. That was the training set. Looking at the manufacturers compromise matrices, the ones derived are not out of the question. I'm working on other training sets, but I'm not expecting to find compromise matrices that provide accurate color for those sets and completely fill the horseshoe, in the case of the cameras used for the above tests.


Jim

[Doug Gray]

Maybe I wasn't clear. The spectral locus for each camera was derived from spectral stimuli 5 nm apart. The gamut boundary was not derived for one set of color patches. That was the training set. Looking at the manufacturers compromise matrices, the ones derived are not out of the question. I'm working on other training sets, but I'm not expecting to find compromise matrices that provide accurate color for those sets and completely fill the horseshoe, in the case of the cameras used for the above tests.


Jim

Jim,

I don't follow. If not from the Colorchecker patches from what did you derive the conversion matrix? I assumed they were derived to minimize some function of dE and then the gamut boundaries are just applying the matrix to the camera curves.

My point was that the Colorchecker, apart from having only 19 patches (the neutrals are effectively redundant), is a limited universe of colors. And even amongst that universe, there are large spectra variations that produce the exact same CIELAB and hence xy chromaticities.

I posit that if one examines other possible spectra for each of the colorchecker patches, without changing their Lab color, that it would be pretty easy to find matrixes that produced a gamut with portions overlapping beyond any arbitrary section of the horseshoe.

Basically, I don't believe the gamut of a camera sensor assy. can be defined by a xy outline of any sort unless the colors and illuminant, from which some sort of "optimal" matrix is generated, are constrained. Of course once any given matrix is generated, the response gamut, using that matrix, is well defined which is what your post shows.

[Jim Kasson]

Jim,

I don't follow. If not from the Colorchecker patches from what did you derive the conversion matrix? I assumed they were derived to minimize some function of dE and then the gamut boundaries are just applying the matrix to the camera curves.

My point was that the Colorchecker, apart from having only 19 patches (the neutrals are effectively redundant), is a limited universe of colors. And even amongst that universe, there are large spectra variations that produce the exact same CIELAB and hence xy chromaticities.

I posit that if one examines other possible spectra for each of the colorchecker patches, without changing their Lab color, that it would be pretty easy to find matrixes that produced a gamut with portions overlapping beyond any arbitrary section of the horseshoe.

Basically, I don't believe the gamut of a camera sensor assy. can be defined by a xy outline of any sort unless the colors and illuminant, from which some sort of "optimal" matrix is generated, are constrained. Of course once any given matrix is generated, the response gamut, using that matrix, is well defined which is what your post shows.

While I think there's discussion to be had about what you're saying about metameric training sets, I think the thrust of what I was saying is captured in your last sentence. Once a compromise matrix is chosen, that matrix, taken together with the sensor/CFA response curves, can limit the colors that you can get to with physical stimuli.

I used to think this was of only academic interest, but my CFA simulation work has made me question this.

More to come, if I can get some time. The a7RIII is about to ship, and I haven't finished testing the D850.

Jim

[Tim Lookingbill]

Has any color scientist produced a photograph of a real scene from a digital sensor that fills the entire spectral locus of human vision?

If so, I'ld like to see it and it better have real objects, not those rainbow looking squares or some Dark Side Of The Moon prism effect.

[Doug Gray]

While I think there's discussion to be had about what you're saying about metameric training sets, I think the thrust of what I was saying is captured in your last sentence. Once a compromise matrix is chosen, that matrix, taken together with the sensor/CFA response curves, can limit the colors that you can get to with physical stimuli.

I agree. That is exactly so once any given matrix is chosen. Another way to look at is that using a single wavelength (or two for the base) and spanning the hue curve some hues will have saturations that are expanded beyond the CIEXY gamut while others will be compressed within it. And there will be hue shifts from the actual along the way.  Using a multiplicity of wavelengths I would expect a different set of expansions/contractions/shifts.

My model of this is not so much a gamut limit as a distortion and, for the multiplicity of wavelengths, a further distortion and spreading of otherwise metameric colors.

It would be interesting to see how much variation occurs when matrixes are chosen based on various sets of real world colors. I would expect they would cohere more closely than using "colors" generated from arbitrary passbands but not as closely as those from the Colorchecker patches.

I used to think this was of only academic interest, but my CFA simulation work has made me question this.

More to come, if I can get some time. The a7RIII is about to ship, and I haven't finished testing the D850.

Jim

Sounds quite interesting Jim. Thanks for the stimulating discussion.

Doug

[Alan Klein]

Has any color scientist produced a photograph of a real scene from a digital sensor that fills the entire spectral locus of human vision?

If so, I'ld like to see it and it better have real objects, not those rainbow looking squares or some Dark Side Of The Moon prism effect.

I've thought about that Tim.  We don't see the entire spectral range either.  Not at the same time.  The eye jumps around in the scene from the light areas to the dark.  Our eye's irises adjust like the aperture on a camera to allow more or less light in as we focus on each of the scene's elements. Then our brain like an HDR program does some algorithm to create an image that is different than any of the individual snapshots our eyes focused on and our brain captured.  It combines all the snaps to make the scene show important content and satisfying details and lighting. A lot of the data is intuited by the brain and never really existed as seen by our eyes.  Even then, the brain does not give equal billing to each of snaps, favoring the lighted areas over the shadow and dark areas. 

It's why when you look at a shot that captured so much range, it looks so drab and we have to add contrast and other changes to make it attractive to our brains aesthetic.  If a camera capture the entire range of what our eyes see over many "shots", we still would have to edit it to make it appear the way our brain sees.  The sensor's range is the least of the issue.  Otherwise, all those slides with limited ranges of 5 stops we use to take beautiful pictures with would not have worked. 

So we humans still have something to say about it. 

[tom b]

On a slightly different note…

I've got a "B&W" painting on my bedroom wall. It's acrylic, which is a bugger to paint with so I added warm and cool paints to adjust the tones. It still looks B&W. So what!

In Photoshop you can do the same thing.

Check it out.

Also Photoshop/Image/Adjustments/Black & White is also worth a look (and is fun to play with).

Cheers,

[Rob C]

Indeed, Tom, and one of the reason's I'd stay with a camera offering all the colours it can.

It's difficult to imagine a situation where more is less when speaking about a single camera; it's gospel with images, but hardly with equipment. But envy (mine) aside, it's one of the things that the new Leica company does very well: sow and encourage the growth of dreams, which, frankly, does no harm at all, and for some, produces the perfect tool as well. It would be divine to be able to pick one up now and again (a rangefinder camera) and just go out and enjoy the differences.

Rob

[Jim Kasson]

Has any color scientist produced a photograph of a real scene from a digital sensor that fills the entire spectral locus of human vision?

Not possible for reflected light, even if the sensor is the eye. Even the gamut of real surface colors has no illuminant that will reach everywhere in the horseshoe.

Even with spectral self-luminous sources, it would be very hard to set up in one shot.

Jim

[Jim Kasson]

It would be interesting to see how much variation occurs when matrixes are chosen based on various sets of real world colors. I would expect they would cohere more closely than using "colors" generated from arbitrary passbands but not as closely as those from the Colorchecker patches.

Sounds quite interesting Jim. Thanks for the stimulating discussion.

I am working with the JPL set of natural spectra, but it's got about 5000 samples, and I need to figure out a way to subset it.

I also have Roy Berns' paint-based set.

And thank you for a discussion that I also found stimulating.

Jim

[digitaldog]

Not possible for reflected light, even if the sensor is the eye. Even the gamut of real surface colors has no illuminant that will reach everywhere in the horseshoe.

And the horseshoe is a theoretical construct.


In the case of the Yxy chromaticity diagram the Spectrum Locus is the path from purple, around the outer edge clockwise, to red (lower left to right corner). Not only does it not contain all the colors visible to humans, it doesn't contain all the hues. The mixing of these hues between purple and red doesn't appear.

Beyond the locus of spectrally pure colors
Mark D. Fairchild*
Munsell Color Science Laboratory, RIT, Rochester, NY, USA 14623-5604

The spectrum locus of a CIE chromaticity diagram defines the boundary within which all physically realizable color stimuli must fall. While that is a physical and mathematical reality that cannot be violated, it is possible to create colors that appear as if they were produced by physically impossible stimuli. This can be accomplished through careful control of the viewing conditions and states of adaptation.

The CIE 1931 xy chromaticity diagram is widely recognized by the horseshoe shape of the spectrum locus. This locus of spectrally pure colors is simply defined by the standard color matching functions, the computation of chromaticity coordinates, and the fundamental limitation of monochromatic stimuli that have energy at a single wavelength and no energy at any other visible wavelength.1 More commonly, the spectrum locus is referred to as defining the boundary of the gamut of all physically possible colors. This, however, is an error. The spectrum locus does define the gamut boundary for physically possible stimuli, but it does not limit the color appearance of stimuli. Specification of color appearance requires more information about the stimulus and adaptation state of the observer2 and it is entirely feasible that a stimulus in one viewing condition might appear to be produced by a stimulus from beyond the locus of spectrally pure colors in some other reference viewing condition.

[Jim Kasson]

In the case of the Yxy chromaticity diagram the Spectrum Locus is the path from purple, around the outer edge clockwise, to red (lower left to right corner). Not only does it not contain all the colors visible to humans, it doesn't contain all the hues. The mixing of these hues between purple and red doesn't appear.

True enough. When the people I know talk about "the horseshoe", they include the straight line linking the points at the spectral positions that close the perimeter so that the area is maximized, so that when you say a triplet is within the horseshoe, you are including all triplets that can be constructed by mixing spectral colors.

Jim

[Jim Kasson]

The spectrum locus does define the gamut boundary for physically possible stimuli, but it does not limit the color appearance of stimuli. Specification of color appearance requires more information about the stimulus and adaptation state of the observer2 and it is entirely feasible that a stimulus in one viewing condition might appear to be produced by a stimulus from beyond the locus of spectrally pure colors in some other reference viewing condition. [/i]

That is undeniable.

Jim

[patjoja]

Leica has a Monochrome camera. No color information, from what I understand.

I have been told that doing B&W this way gives a higher quality image file for B&W.

Is this correct?

Is there an actual advantage to a fully monochrome sensor? If so, does it react to filtration as film does? So I can actually choose a wratten 23/25/29 and similar rather than rely on nebulous sliders in post processing programs?

Sure would like to see a Fuji X-Pro2M (monochrome) as I sure can't afford the Leica gear.

Astrophotographers have been using monochrome sensors plus filters for quite some time.  The reason is increased resolution and sensitivity compared to a bayer matrix.  I'm not sure I would do it myself for daytime photography though given the costs and excellent quality of cameras these days.  They are certainly a niche camera, but YMMV.

Patrick