Can a prism deliver all the colours of a rainbow?

[Mike Brennan]

A general question,

 

Is it possible for a prism with 3 filtered ccds to deliver all the colours of the rainbow?

Forget the A/D for the moment and the display too.

 

Is there any intrinsic physics/optic properties of a prism and filtered ccd imaging device that limits or reduces the colour pallet?

 

 

 

 

Mike Brennan

[Phil Rhodes]

Hi,

 

The gamut is limited by the sensitivity of the CCDs. Despite the monosensitive chips and the monochromatic image falling on each of them, it's still red, green or blue light and the CCD will react to it as per its frequency response.

 

Furthermore, the filters aren't perfect. The red channel sees some blue and green light. This reduces the effective saturation, but it's an extremely minor issue with modern dichroic filters. There's also some wavelengths that the filters might never pass, so there may be some colours you just don't see. Equally tiny issue.

 

In the main, the problem is with the chips seeing more than human vision in the UV and infra-red, requiring filtering for both.

 

Phil

[David Mullen ASC]

The prism itself is just used to split the light three ways, not separate wavelengths. The filters are what cancel some wavelengths, allowing a record of specific red, blue, or green info. So besides the accuracy of the filters in terms of how perfectly they cancel light, I guess the real issue is whether you believe that all colors can be accurately represented by combinations of red, blue, and green.

 

Imperfect filters that still allow a little of the other two colors to pass through mainly just lower saturation, not accuracy.

 

The whole system is very similar to that of a 3-strip Technicolor camera, which had very good color reproduction. But you can't really separate the camera system from how that information is stored and reproduced.

 

Also, Phil is right that the problem is also sensitivity to wavelengths the human eye DOESN'T see.

[John Sprung]

The prism passes all the colors, but you lose gammut when they hit the filters and CCD's. What you want to learn about is the CIE 1931 chromaticity diagram:

 

http://www.yorku.ca/eye/ciediag1.htm

 

That sort of shark fin shaped graph ranges from pure single wavelengths around the perimeter of the curve to the various things we call white near the middle. Color temperatures fall on a small curve inside the white region. The straight line connecting the ends of the big curve are the red-blue mixture colors.

 

Any three color film or video system has three primaries that form a triangle on the diagram. The system can reproduce colors that lie inside that triangle, but not those that lie outside it. For us the good news is that the colors that lie outside the gammut of our film and video systems are only those that are so extremely saturated that we hardly ever encounter them in the real world. Subtle colors like skin tones can go wrong on us, but not because we don't have enough color gammut.

 

 

 

-- J.S.

[John Pytlak RIP]

Kodak Color Scientist Dr. David MacAdam was one of the pioneers in color science and color specification within the CIE coordinate system:

 

http://home.wanadoo.nl/paulschils/08.02.html

 

http://www.colorpro.com/info/data/cie.html

 

I was very fortunate to have taken a color science course from Professor MacAdam at the University of Rochester, where he taught after retiring from Kodak.

 

Tutorial on color measurement and specification:

 

http://www.handprint.com/HP/WCL/color6.html

[Guest Daniel J. Ashley-Smith]

It depends what bit the CCD's are.

 

The prism itself is just used to split the light three ways

So the prism splits the colours up? How does it do that?

Edited November 4, 2004 by Daniel J. Ashley-Smith

[Phil Rhodes]

Hi,

 

It's a glass block designed to split the light equally three ways, using internal mirrors, and it has dichroic filters coated onto some of the surfaces. They're quite pretty.

 

Phil

[Guest Daniel J. Ashley-Smith]

I see, so the light is split up into the 3 ways (or depending on how many CCD's there are), and then using the dichroic filters, the unwanted colours are removed accordingly. So the CCD's recieve only the correct colours.

 

Am I on the right tracks?

 

(On the odd note, how is dichroic pronounced "Die-crow-ic"? Just don't want to go round saying it wrong and looking like a complete prat)

Edited November 4, 2004 by Daniel J. Ashley-Smith

[John Pytlak RIP]

I see, so the light is split up into the 3 ways (or depending on how many CCD's there are), and then using the dichroic filters, the unwanted colours are removed accordingly. So the CCD's recieve only the correct colours.

 

Am I on the right tracks?

 

(On the odd note, how is dichroic pronounced "Die-crow-ic"? Just don't want to go round saying it wrong and looking like a complete prat)

<{POST_SNAPBACK}>

 

Kodak has a good on-line course about color:

 

http://www.kodak.com/US/en/digital/dlc/boo...er2/index.shtml

 

And digital photography:

 

http://www.kodak.com/US/en/digital/dlc/index.jhtml

 

And lots about motion-picture technology:

 

http://www.kodak.com/country/US/en/motion/.../h1/index.shtml

 

http://www.kodak.com/US/en/motion/support/h2/

 

http://www.kodak.com/US/en/motion/support/...d=0.1.4.9&lc=en

 

http://www.kodak.com/US/en/motion/support/...0.1.4.9.6&lc=en

Edited November 5, 2004 by John_P_Pytlak

[Balazs Rozsa]

I see, so the light is split up into the 3 ways (or depending on how many CCD's there are), and then using the dichroic filters, the unwanted colours are removed accordingly. So the CCD's recieve only the correct colours.

 

Am I on the right tracks?

<{POST_SNAPBACK}>

 

The light is not split up into 3 ways first and then filtered for each of the CCDs, but the splitting and the filtering happens at the same time. For the red CCD the red portion of the light is extracted from the light beam and directed towards the red CCD, for the blue CCD the blue light is extracted and so on.

 

This may not be a big difference, but in my opinion this is the nice thing about a prism video camera: all the incoming red light reaches the CCD pixels. If the light was split into 3 ways and then filtered, only about one third of the red light would reach the red CCD. The same is true for the bayern filtered one CCD cameras: only about one third of the light gets to the pixels, the rest is blocked in the filters. This is the reason 3 CCD cameras are more sensitive to light than 1 CCD cameras.

 

At http://www.perrybits.co.uk under Articles in "Article 4: How does a video camera capture color?" you can read a precise desription how the prism and the filters work.

 

Balazs

[Mike Brennan]

OK so 3 chip ccd block can capture a very good range of colours.

In my view skin tone lcks the range of tonality we see with the naked eye and that is recorded on film.

 

Perhaps the problem with the colour pallette (range of colours) with digital video, *when it comes to reproducing skin tones* is the number of bits in the A/D.

 

Viper, f900 have 12bits new model of Genesis will have 14bit AD.

Thomson have a 16bit AD in SD camera.

High end digital stills cameras have 16bit AD and most closely resemble nature (and the result when using film)

 

Assuming 4:2:2 sampling, would a 16bit AD deliver more natural skin tone? (for film out)

 

Has Kodak loaded or emphasised skin tone response in its film stocks over the years?

Would a tweake, maybe even user adjustable be an idea for digital cameras, to load more bits into the skin colour area of the spectrum??

 

 

Commments??

 

 

Mike Brennan

[Phil Rhodes]

Hi,

 

> OK so 3 chip ccd block can capture a very good range of colours.

> In my view skin tone lcks the range of tonality we see with the naked eye and

> that is recorded on film.

 

I find it difficult to make that comparison because skin tones recorded on film look so completely different to the way they look with the naked eye.

 

> Perhaps the problem with the colour pallette (range of colours) with digital

> video, *when it comes to reproducing skin tones* is the number of bits in the

> A/D.

 

I don't believe this is true. Film stuff still looks nicer when transferred to video, and no video format has 12 or more bits of quantisation. It's not an A/D resolution issue. In any case, many of those bits will be noise, irrespective of the origination format.

 

> High end digital stills cameras have 16bit AD and most closely resemble nature

> (and the result when using film)

 

I'm still not sure this is anything to do with the A/D resolution. The images are still clearly better when viewed on an 8-bit monitor. Consider also that DVX-100 hack. The DVX-100 is a respectable handycam but holds no torch to a Viper, or most digital stills cameras. Take the data directly off its (12-bit) A/D chips, before it gets processed, and "better" is hardly the word - it looks, spookily, like a slightly low-res digital still camera shot and has over a stop of extra dynamic range. It seems to me that a lot of video cameras do very deeply inadvisable things to the image in DSP processing, probably to make them look more sensitive than they really are, and this is where the problem lies.

 

> Assuming 4:2:2 sampling, would a 16bit AD deliver more natural skin tone? (for

> film out)

 

No, since it would almost certainly not be recorded to a video tape format as 8- or 10-bit. Record it as 10bit log to a hard disk array and you might see improvements, but not necessarily in the way skintones are rendered. I'm not really sure what you mean by the way skin tones are rendered - it's possible to shoot perfectly adequate beauty closeups on video if you know what you're doing -but it's not about raw resolution.

 

Phil

[Mike Brennan]

Hi,

 

> Assuming 4:2:2 sampling, would a 16bit AD deliver more natural skin tone? (for

> film out)

 

No, since it would almost certainly not be recorded to a video tape format as 8- or 10-bit. Record it as 10bit log to a hard disk array and you might see improvements, but not necessarily in the way skintones are rendered.

 

Phil

<{POST_SNAPBACK}>

 

Yes I understand that we drop back the number of colours when we record 12 bit AD to 10 bit

But one should not assume that a 10bit AD to 10 bit recording has same exactly same range of colours as 12bit AD to 10bit recording.

 

By "render" I mean just that, film packs more tones into skin than current video cameras. 16 bit digital still cameras come closest. Ikegami and Sony seem to be making 14bit AD as standard on next generation cameras.

 

I figure it is the number of bits assigned to AD and the effect is not lost on 8 bit or 10 bit recording. (Head shots originated on film and transferred to HDCAM have more tones less plastic to the pro eye at least, than f900. I'm excluding grain from this subjective evaluation)

 

Add more bits to the AD as on large format ditial stills cameras and digital renders skin tones more like film!

 

Here is a link to digital pics shot with a 16 bit AD in the camera.

http://www.creo.com/global/products/digita...af/default.htm#

 

 

 

Mike Brennan

[John Pytlak RIP]

Has Kodak loaded or emphasised skin tone response in its film stocks over the years?

Would a tweake, maybe even user adjustable be an idea for digital cameras, to load more bits into the skin colour area of the spectrum??

Commments??

Mike Brennan

<{POST_SNAPBACK}>

 

Today's excellent flesh tone reproduction of color negative films has been an evolutionary process. It really starts out with a good, neutral tone scale, and then optimizing the color sensitivity and color interimage to give the most natural color and flesh tone reproduction.

 

Kodak image scientist David Long has a very good Q&A about the Kodak VISION2 technolgies on the Kodak website:

 

http://www.kodak.com/country/US/en/motion/...s/v2/long.shtml

[Phil Rhodes]

Hi,

 

> Head shots originated on film and transferred to HDCAM have more tones less

> plastic to the pro eye at least, than f900

 

That's the thing - they actually don't, technically speaking. You're objecting to the way the CCD is seeing it. Higher-bit ADCs can make for cleaner images or more range for adjustment before contouring becomes an issue, but that's about it. Once again I think a lot of this has to do with the camcorder manufacturers' insistence on massively amplifying (6-8dB) the picture to increase apparent sensitivity. Great to have the option but they seem to quite intentionally chop dynamic range enormously.

 

> Add more bits to the AD as on large format ditial stills cameras and digital

> renders skin tones more like film!

 

As I said - so does a DVX-100 if you pull out all the DSP. Look at www.reel-stream.com

 

Phil

[Mike Brennan]

> Head shots originated on film and transferred to HDCAM have more tones less

> plastic to the pro eye at least, than f900

 

Phil replied

<<That's the thing - they actually don't, technically speaking.>>

 

I disagree. Skin tones shot on film and transferred to HD look better than if they originated on f900.

 

Some 4:4:4 HD vs film camera tests that are transferred to film carefully choose subject matter to minimise the difference. So sometimes skin tones can appear similar but most often not.

The diference is not something the audience cares that much about by the way.

 

<<You're objecting to the way the CCD is seeing it. Higher-bit ADCs can make for cleaner images or more range for adjustment before contouring becomes an issue, but that's about it.>>>

 

Since there are so few 14bit ADs in video cams I'm guessing, but given the evidence from digital stills cameras with 14 and 16bit ADs then I expect an improvement in tonal range with 14bit ADs.

 

<<<Once again I think a lot of this has to do with the camcorder manufacturers' insistence on massively amplifying (6-8dB) the picture to increase apparent sensitivity. Great to have the option but they seem to quite intentionally chop dynamic range enormously.>>>>>

Sounds like someone should come up with a Hack, -3db doesn't do much.

 

I wrote

> Add more bits to the AD as on large format digtial stills cameras and digital

> renders skin tones more like film!

Hil wrote

<<<As I said - so does a DVX-100 if you pull out all the DSP. Look at www.reel-stream.com>>>

 

This reel-stream site is just a 444 recording from a 3 chip ccd camera with 10 bit AD . A 444 recording from f950 or Viper (with 12bit AD) is better but still not up to film in respect of skin tones. Watch the difference in the skin tones in Collateral between film and Viper

 

 

The difference isn't huge but something worth aiming for!

 

Would be cool if the young guns at reel-stream could hack 4:4:4 from a f900!

 

 

Mike Brennan

[Phil Rhodes]

Hi,

 

> Skin tones shot on film and transferred to HD look better than if they originated

> on f900.

 

I'm not necessarily disagreeing with you, that may be true - I don't think it necessarily is for all circumstances, but that's a matter of taste - but it's not do with bit count. It might possibly be to do with linearity and DCC interpolation, CCD sensiometry and lighting, but I don't think it's to do with bit count because cameras with widely varying bitcounts display the same strengths and weaknesses.

 

> This reel-stream site is just a 444 recording from a 3 chip ccd camera with 10

> bit AD

 

12, actually, but that's not really the point. I was trying to illustrate the way a motion video CCD can start looking a lot more like a digital stills camera if you take out some of the processing. Ergo:

 

> A 444 recording from f950 or Viper (with 12bit AD) is better but still not up to

> film in respect of skin tones

 

Yes exactly - and such a recording is made AFTER the onboard DSP processing so it still looks rough. I don't think it's an A/D thing, I think it's a DSP processing thing. The reel-stream output does look a lot more like a digital stills camera and that's what I find interesting about it. I'd very much like to see a direct takeoff of the DSP data from an F900 done in the same way, I'm sure it would look much better - and isn't this basically what a Viper does?

 

Phil

[Mike Brennan]

Hi,

 

 

Yes exactly - and such a recording is made AFTER the onboard DSP processing so it still looks rough. I don't think it's an A/D thing, I think it's a DSP processing thing. The reel-stream output does look a lot more like a digital stills camera and that's what I find interesting about it. I'd very much like to see a direct takeoff of the DSP data from an F900 done in the same way, I'm sure it would look much better - and isn't this basically what a Viper does?

 

Phil

<{POST_SNAPBACK}>

 

I think we have got to the nub of the question, it is worth clarifying what in-camera processing takes place on the reel-stream and the Viper f950 when outputting 4:4:4

 

 

 

Also Charles Poynton says that Y CbCr has very poor codeword utilisation with only 17% of the codewords representing colours.

 

Can someone put this in context?

 

Is this where the missing colours go...?

 

 

 

Mike Brennan

[Mike Brennan]

"Of the 16.7 million colors available in studio R?G?B?, only about 2.75 million are available in Y?CBCR. If R?G?B? is transcoded to Y?CBCR, then transcoded back to R?G?B?, the resulting R?G?B? can?t have any more than 2.75 million colors! 4 220 2784375 10648000 0261 3 ==.

 

I haven't seen RGB to HDTV described in this way before.

Here is the white paper from Charles Poynton

 

"http://www1.discreet.com/SITE/whitepapers.nsf/0ce70f9723c01b7485256562005e6153/b6f5ed7125054754852565ef00743e26/$FILE/Merging_RGB_and_422.pdf

 

 

Are we saying 4:2:2 can only register 2.75 million colours from a CCD?

 

Mike Brennan

[Phil Rhodes]

Hi,

 

> I think we have got to the nub of the question, it is worth clarifying what in

> -camera processing takes place on the reel-stream and the Viper f950 when

> outputting 4:4:4

 

As little as possible - just enough for there to be an output amp on the CCD and proper bias adjustment, then the ADC.

 

> Also Charles Poynton says that Y CbCr has very poor codeword utilisation with

> only 17% of the codewords representing colours.

> Can someone put this in context?

 

Sorta. The problem is that the RGB colour space captured by the CCD does not overlap 1:1 with the YUV colourspace recorded to tape. Exactly what loss takes place at this point will depend on a lot of things, including the content of the frame and the camera settings.

 

Phil