The look of a CCD versus a look of the CMOs in medium format

[eronald]

There is an interesting online color test which people could take before they explain how important precise color is to them

I believe there are real skin and texture issues with CMOS implementations, maybe it is signal processing. Color seems ok; on color alone I would agree with the other techies here that color is mostly a CFA and Silicon process issue.

So I would expect landscapers and still-life studio shooters to be happy with the new CMOS, if they can get it in full frame, the fashion, beauty and portrait guys may want to stay with the older, smaller microlensed CCDs eg P30+, Leica S etc.

Edmund

PS an interesting test of texture would be to shoot a monochrome CCD against a monochrome CMOS

[tho_mas]

There are a few technical differences that may lead to some differences, such as higher dynamic range of recent CMOS technology, and a higher sensitivity for color cast at extreme angles of incidence on current CMOS implementations, but not something that would define a 'look', like the difference between Fujifilm Provia and Astia.

CMOS sensors allow on-chip pre-processing. Sony clearly states that they perform a first stage of Noise Reduction on chip (and of course this can't be undone in post production).
IMO this may of course influence the "look"... don't you think so?

[eronald]

CMOS sensors allow on-chip pre-processing. Sony clearly states that they perform a first stage of Noise Reduction on chip (and of course this can't be undone in post production).
IMO this may of course influence the "look"... don't you think so?

look and think in the same sentence? You do expect a lot of your fellow man

Edmund

[Bart_van_der_Wolf]

There is an interesting online color test which people could take before they explain how important precise color is to them

Hi Edmund,

True, but there is a lot more to the issue than objectively ranking color with precision. FWIW, I had a zero error score when I did the test earlier ...

I believe there are real skin and texture issues with CMOS implementations, maybe it is signal processing. Color seems ok; on color alone I would agree with the other techies here that color is mostly a CFA and Silicon process issue.

Given the specific color reflection (specular, reflected ambient, and diffuse) of human skin, and its high degree of diffuse Infra-red reflection from the deeper skin layers, mixed with a more or less superficial (IR absorbing) vein structure that contracts and dilates with temperature and emotion, and adding a dose of variable pigmentation, there is not a single simple skin color, not even for an individual ... Yet, most skin reflection does fall in a relatively narrow range of Hue angles, which is not influenced by IR sensitivity of the sensor which mostly just produces a different saturation because of the transparency of the CFA filters for IR. All silicon based sensors are sensitive to Near-IR from 700-1000 nm, unless the camera filters that out before it hits the sensor.

So we are dealing with an IR-filtration, and CFA tuning issue that makes one design more suitable for skin rendition than another design. Add a Raw converter's profiling accuracy for a given illuminant, and we can say that these are the defining factors, not the photon to electron conversion technology, AKA CCD or CMOS device.

But the OP doesn't want to discuss technology, so I'll leave it at that.

Cheers,
Bart

[eronald]

Bart,

 skin *texture* is more than color. texture is not yet well understood mathematically, which is why renderings are not specced.
 i think one reason people liked film was that skin texture rendered well, and also film grain blended naturally into texture where it became apparent. Of course, film was tuned over a century to be pleasing - people who made ugly films went broke

Edmund

Hi Edmund,

True, but there is a lot more to the issue than objectively ranking color with precision. FWIW, I had a zero error score when I did the test earlier ...

Given the specific color reflection (specular, reflected ambient, and diffuse) of human skin, and its high degree of diffuse Infra-red reflection from the deeper skin layers, mixed with a more or less superficial (IR absorbing) vein structure that contracts and dilates with temperature and emotion, and adding a dose of variable pigmentation, there is not a single simple skin color, not even for an individual ... Yet, most skin reflection does fall in a relatively narrow range of Hue angles, which is not influenced by IR sensitivity of the sensor which mostly just produces a different saturation because of the transparency of the CFA filters for IR. All silicon based sensors are sensitive to Near-IR from 700-1000 nm, unless the camera filters that out before it hits the sensor.

So we are dealing with an IR-filtration, and CFA tuning issue that makes one design more suitable for skin rendition than another design. Add a Raw converter's profiling accuracy for a given illuminant, and we can say that these are the defining factors, not the photon to electron conversion technology, AKA CCD or CMOS device.

But the OP doesn't want to discuss technology, so I'll leave it at that.

Cheers,
Bart

[Bart_van_der_Wolf]

CMOS sensors allow on-chip pre-processing. Sony clearly states that they perform a first stage of Noise Reduction on chip (and of course this can't be undone in post production).
IMO this may of course influence the "look"... don't you think so?

Hi Thomas,

Depends on the particular Noise reduction at hand, which is not the kind we do with tools like Topaz Denoise after demosaicing. The noise reduction is probably a reduction of per sensel read noise, dark current, and calibration for pattern noise suppression. Nothing to do with noise that propagates through demosaicing into low spatial frequency color blobs.

Besides, CMOS keeps a lot cooler than CCDs because they require less power to operate, which saves batteries and reduces temporal noise build up, making them usually more suitable for long exposure time scenarios. IOW, they can be used to shoot certain scenes that a CCD would have more difficulty with. Horses for courses.

Cheers,
Bart

[Bart_van_der_Wolf]

skin *texture* is more than color. texture is not yet well understood mathematically, which is why renderings are not specced.

Sure, and the skin of a single person also exhibits multiple different textures across the face and body, which can be characterized by specular reflectivity (which is influenced by skin grease, pore structure, and elasticity/alignment of subcutaneous tissue). Skin is a very complex 'material' to photograph, so some gear may be tuned to do it better than other gear.

Many people say that they like the tuning for skin reproduction of Leaf backs (CCD), better than Phase One backs (CCD), which proves it has little to do with the CCD vs CMOS myth.

Cheers,
Bart

[ErikKaffehr]

Hi,

Some reflections on skin:

- Skin is often masked by make up in most commercial shots, so I don't think we can talk about natural skin tones.

- It may be that aliasing can add some artificial structure to skin. Dan Browning had a good posting on that back in 2011, but those links are no longer valid.

I am more of a landscape shooter, so I cannot talk about skin.

I would also say that we have two many variables:

- CCD vendors are Kodak and Dalsa
- CMOS vendors are Sony and some else

Than we have raw processing, some use Capture One with Phase backs and some uses Canon's with Lightroom/ACR and some use something else. ACR rendition is much governed by Thomas Knoll, while C1 rendition depends on the "Image Professor".

To many variables. Now, we have the Phase One IQ series, IQ250 CMOS and IQ260 and IQ280 on CCD. Lets wait for a good comparison…

Best regards
Erik

Sure, and the skin of a single person also exhibits multiple different textures across the face and body, which can be characterized by specular reflectivity (which is influenced by skin grease, pore structure, and elasticity/alignment of subcutaneous tissue). Skin is a very complex 'material' to photograph, so some gear may be tuned to do it better than other gear.

Many people say that they like the tuning for skin reproduction of Leaf backs (CCD), better than Phase One backs (CCD), which proves it has little to do with the CCD vs CMOS myth.

Cheers,
Bart

[hjulenissen]

skin *texture* is more than color. texture is not yet well understood mathematically, which is why renderings are not specced.

I'd suggest that what people is calling "texture" may not yet well understood _perceptually_. Once it has been unambigously described by those using the term, the mathematical part be trivial, and may or may not be a simple transformation of something like MTF measurements.

-h

[Bart_van_der_Wolf]

I'd suggest that what people is calling "texture" may not yet well understood _perceptually_. Once it has been unambigously described by those using the term, the mathematical part be trivial, and may or may not be a simple transformation of something like MTF measurements.

With apologies to the OP for being off topic a bit (although very much related to 'the look'):
Phong lighting/shading.

Cheers,
Bart

[eronald]

With apologies to the OP for being off topic a bit (although very much related to 'the look'):
Phong lighting/shading.

Cheers,
Bart

Bart,

  skin, cloth, plants, rocks are not surfaces. They are textures.
  if you prefer, they have a complex, fractal related microstructure, which is not quite scale invariant.
  some have embedded microspeculars.

 I'm tired saying the same thing. If we knew how we perceive these things,  we'd have rendering criteria, and we wouldn't have camera issues.

Edmund

[tho_mas]

The noise reduction is probably a reduction of per sensel read noise, dark current, and calibration for pattern noise suppression.

"probably" ... but maybe there is more than that. And maybe there is even more pre-processing going on than just some kind of Noise Reduction. I don't know. But on-chip processing might be a factor that CMOS captures in fact sometimes look somewhat flat and "digital".

Besides, CMOS keeps a lot cooler than CCDs because they require less power to operate, which saves batteries and reduces temporal noise build up, making them usually more suitable for long exposure time scenarios.

is there any CMOS based camera that shoots as long and as clean as a P45+?

[ErikKaffehr]

Hi,

In Swedish we have a proverb, "sila mygg och svälja elefanter", according to Google translator it would be: "straining at gnats and swallowing elephants".

Anyway, a program like Capture One does pretty hefty noise reduction as default. If you process any image in C1 with default settings, pretty aggressive noise reduction will be applied and that also applies to sharpening, both are probably far more intensive than whatever Sony may do on chip. Yes, LR 5/ACR has also noise reduction as default, but probably less than C1.

Best regards
Erik

"probably" ... but maybe there is more than that. And maybe there is even more pre-processing going on than just some kind of Noise Reduction. I don't know. But on-chip processing might be a factor that CMOS captures in fact sometimes look somewhat flat and "digital".
is there any CMOS based camera that shoots as long and as clean as a P45+?

[Bart_van_der_Wolf]

is there any CMOS based camera that shoots as long and as clean as a P45+?

Would be an interesting comparison, but one can downsample a CMOS image to one quarter of the number of sensels and reduce perhaps 50% of the noise of a 1 hour exposure, or use 4 exposures of 15 minutes, add them, and then downsample to 25% of the total number of sensels.  The downsampling won't be as effective as direct binning before read-out, but the read noise would already be much lower to begin with. All CMOS devices should be able to do that, if the camera has a Bulb setting for exposure.

Cheers,
Bart

[EricWHiss]

There was a different look between my Kodak sensor based digital backs and my dalsa sensor digital backs and both are CCD.  There are a lot of people saying the CMOS and CCD chips have got to be the same from some theoretical point of view - maybe, but in practice I don't find that IMHO.  I have two monitors on my setup a NEC and an Apple and no matter how hard I work with the spectra view calibration, I can't make them have the same look.  I can match the colors mostly, but the look is still different.

Getting back to cameras and CCD vs CMOS - its a complete system - not just the sensor chip but all the on board processing plus the rest - and of course even the lenses can affect color and look.  

I did lots of tests with the d800e vs my AFi-ii 12 and while the D800e detail in the luminous channel is very good, the color is flat as if it were averaged.   The color tonality does impact the look and is still an area where I think CCD chips excel.  Does it have to be that way? No, probably they could be made more similar but every designer using the CMOS seems to be looking for higher ISO performance instead of optimizing IQ at base.  The CFA choices and other on board processing probably account for some of the differences in look (without talking about lenses). I suspect whatever profiles the manufacturer use also impact the look.

[ErikKaffehr]

Hi,

P45+ does no binning on sensor. Some of the other PXX+ backs do.

I am not really in the long exposure business, but Ray "ondebanks" asked for a set of long exposures from my P45+, which I supplied to him. I am not really sure what he made of it.

Best regards
Erik

Would be an interesting comparison, but one can downsample a CMOS image to one quarter of the number of sensels and reduce perhaps 50% of the noise of a 1 hour exposure, or use 4 exposures of 15 minutes, add them, and then downsample to 25% of the total number of sensels.  The downsampling won't be as effective as direct binning before read-out, but the read noise would already be much lower to begin with. All CMOS devices should be able to do that, if the camera has a Bulb setting for exposure.

Cheers,
Bart

[tho_mas]

Anyway, a program like Capture One does pretty hefty noise reduction as default. If you process any image in C1 with default settings, pretty aggressive noise reduction will be applied and that also applies to sharpening, both are probably far more intensive than whatever Sony may do on chip.

Software based post processing is a different story. My default for luminous NR in C1 is zero (for all my cameras) and pretty low values for color NR.
The point is: on-chip pre-processing can't be reversed.

[Fine_Art]

Sure, and the skin of a single person also exhibits multiple different textures across the face and body, which can be characterized by specular reflectivity (which is influenced by skin grease, pore structure, and elasticity/alignment of subcutaneous tissue). Skin is a very complex 'material' to photograph, so some gear may be tuned to do it better than other gear.

Many people say that they like the tuning for skin reproduction of Leaf backs (CCD), better than Phase One backs (CCD), which proves it has little to do with the CCD vs CMOS myth.

Cheers,
Bart

IMO capturing skin well is more about fine contrast than color. A given small area, ignoring veins, probably has very similar color. The main difference is when light hits a small crease it looks flat. When it hits the high smooth sections it may look shiny. So the color fractaling is a byproduct of the angle relative to the incoming light. It's the same on the surface of a leather sofa. If you look straight at the reflection of a bulb the texture seems to have huge contrast. If you look where light is all bouncing away it looks like light brown lines around dark brown spots.

The issue for the camera is how close is this detail to the limits of the lens/ sensor? Most portraits put fine lines at a frequency close to being extinguished just based on the camera resolution vs the need for some room around the person's head. If someone does a 4 shot stitch of a face portrait we may find the skin 'issues' go away.

[Paul2660]

The P45+ does a very nice job on the longer shots.  I took mine to 55 minutes working at night. Images had very low noise and no excessive noise reduction in C1.  All at iso 50.

I have yet to shoot the 260 at the 140 iso setting but from what I seen from other photographers it does not seem at noise free as the P45+.

Paul

[Hywel]

Hi,

Some reflections on skin:

- Skin is often masked by make up in most commercial shots, so I don't think we can talk about natural skin tones.

- It may be that aliasing can add some artificial structure to skin.

That is certainly my experience. I shoot with a Hasseblad H3D31ii (microlensed). Aliasing often shows up a "rainbow" speckling on skin micro-highlights where pores are hit by hard light, for example.

I absolutely LOVE the resulting shimmer- it is like applying a very, very fine subtle rainbow glitter makeup with an airbrush. (Which we've tried, and which is a damn sight harder to create and photograph for real). 

Even when sent through a skin smoothing pass and downsampled in post, an impression of the shimmer remains - contributing to what others have called the skin texture.

This is an out-and-out technical ERROR. This colour detail is aliased, not real.

Allied with the extra apparent sharpness and micro-contrast that shooting without an anti-alising filter makes, plus the post-production diffusion techniques which artificially preserve the fine detail compared with the way you'd record it with an AA filter or on-lens diffusion, the net result is my preferred rendition of the skin of pretty girls. (Which is 98% of my professional photography).

The sum of errors and imperfections can often get you to a more interesting creative place than a technically accurate rendition.

More than anything it is the way the camera renders skin texture that makes me choose it over my Canon when I have enough light to work with. The Canon gives nice enough results, just not the "fairy dust turned up to 11" look that I particularly like.

By contrast, to get the look I like best from motion photography I tend to use diffusion and smoke and blurry old stills lenses when shoot motion- I think the Hasselblad "sparkle" would probably look ghastly with shifting aliasing in motion. It just looks stunning in a single frame

The only answer is, as always, use the tool that gets you the result you want. And hire before you buy!

  Cheers, Hywel