beejaybee
Marvin
Reged: 18/07/2007
Posts: 4494
Loc: Really Here In Name Only
|
|
Just thinking - why does the Bayer matrix use red green & blue filters? If it used cyan magenta & yellow (i.e. minus red, minus green and minus blue) twice as many photons would be captured, leading to significantly less image noise. The RGB values traditionally used in digital signal processing could of course be regenerated trivially.
|
LargeFormat
old hand
Reged: 24/10/2006
Posts: 1066
Loc: Buckinghamshire and Cumbria
|
|
Having just experimented with RGB and CMY I wonder if loss of contrast may be the problem? CMY is used additively as in printing. Although one's photographic file is RGB the printer uses CMY plus K (black) to print. (I know BJB knows this) There is clearly scope for other arrangements as shown by Kodak's matrix proposal to achieve a similar end.
|
beejaybee
Marvin
Reged: 18/07/2007
Posts: 4494
Loc: Really Here In Name Only
|
|
Quote:
Having just experimented with RGB and CMY I wonder if loss of contrast may be the problem?
This may be the case when the CMY are dyes as used in printing. But, for filters, it is possible to make bandpass filters with very high transmission in band, very low transmission out of band and very sharp transitions. Interference filters are not cheap to make but the filter plate required for a sensor is not especially large. I think lots of users would be prepared to pay say £50 extra for a camera with twice the sensitivity or half the image noise for the same size & resolution sensor in a DSLR, maybe only £5 or £10 extra for a small sensor for a compact camera.
In fact the RGB filters used in Bayer matrices seem to be dye based, so their in band transmission is lower than optimal and the cutoffs are rather weak. Switching to an interference based CMY filter matrix could very well result in a quantum efficiency gain considerably greater than two, and might actually improve contrast and the rendition of hues and saturation rather than the reverse.
|
huwevans
Old Hand
Reged: 05/08/2000
Posts: 15272
Loc: Dorset, UK
|
|
I don't know, but a couple of things occur to me.
Firstly, there has been some experimenting with different CFA types over the years. The one that rings bells in this context was Sony's RGBE (Emerald) design of a few years ago. It didn't (as far as I could see) give any obvious advantage though, and didn't take on. I'm not sure if Sony still use it at all.
But anyway, it's a fairly obvious idea to try different basis sets for the colour space, so I presume there must be some good reason why we still use RGB arrays, more than thirty years after their invention.
Secondly, I'm wondering if a CMY array wouldn't perhaps mean that the sensor was only poorly able to replicate the way that we see with our eyes. I'm thinking here about the ability to discriminate between subtle variations of particular shades.
AIUI the spectral sensitivity of the human eye has distinct peaks that roughly correspond to the additive primaries, red, green, and blue. Indeed, that is why we consider them to be additive primaries - it's not actually anything to do with the properties of the light itself. And thus our ability to discriminate between different shades of, say, yellow is poorer than that for red or green. Anyway, I'm just wondering whether that would mean that an RGB sensor is likely to give the best colour discrimination in those bands where our eyes also work best. And that similarly a CMY sensor might discriminate colour most efficiently in the wrong bands.
You could thus certainly have a usable system based on a CMY filter array, but it wouldn't see colour quite the way we do. And although it wouldn't produce weird effcts like cross-processing, or whatever, it probably would look subtlely different, and maybe not quite right in some hard to describe way. Taking things to the logical extreme, of course, you wouldn't need three colours at all - two would be sufficient, if the shape of the filters was right. But again I don't think the pictures produced would look to us quite the way the original scene did.
--------------------
Huw Evans.
  
|
Terrywoodenpic
A whiff of silicon...
Reged: 21/01/2006
Posts: 285
Loc: Saddleworth UK
|
|
I have wondered for some while why something like stochastic screening was not used, it uses extremely tiny dots randomly scattered across the surface, similar to the grain in photographic film, to produce rich colour halftones with exquisite detail.
--------------------
63 happy photo years from amateur to professional and back to amateur
|
beejaybee
Marvin
Reged: 18/07/2007
Posts: 4494
Loc: Really Here In Name Only
|
|
Quote:
I'm wondering if a CMY array wouldn't perhaps mean that the sensor was only poorly able to replicate the way that we see with our eyes. I'm thinking here about the ability to discriminate between subtle variations of particular shades.
I don't see the problem here.
If we use interference filtering we can make the cyan filter have a "double humped" bandpass which exactly mirrors the superposition of the existing blue and green filters, etc. Having done this we just do the usual processing:
Red = Magenta + Yellow - Cyan
Green = Yellow + Cyan - Magenta
Blue = Cyan + Magenta - Yellow
and we're back to exactly the same spectral response as we started off with. But we still captured twice as many photons with the same intensity of light hitting the sensor for the same amount of time, which implies significantly less noise, at least in the luminance channel which is the one where our vision is most sensitive.
Of course we may wish to vary the spectral response subtly ... the existing RGB filters are not perfectly tuned to the (normal) human trichromatic vision system, which (for instance) has a deeper dip in the response between red and green than the Bayer RGB filter set does.
|
huwevans
Old Hand
Reged: 05/08/2000
Posts: 15272
Loc: Dorset, UK
|
|
You're confusing me now - the whole point of your CMY array depends on the filters being band-stop, not band-pass, doesn't it? If you're just talking about band-pass filters then the situation is surely just analogous to the current RGB filtration, except for the filters being centred on a different set of frequencies - where's the advantage? Or what am I missing?
--------------------
Huw Evans.
|
OneTen
'Two Breakfasts'
Reged: 23/06/2003
Posts: 2471
Loc: Devon
|
|
New keyboard required...because my head has just exploded.  
--------------------
Richard .......... My Website - My Blog - My Flickr
|
daft_biker
Action Man!
Reged: 11/10/2006
Posts: 7226
Loc: Doon the glen
|
|
If twice as many photons are captured would highlights blow sooner?...unless you half the exposure and then you're back to square one? 
--------------------
Andrew (BSRIPN) ... Pics.
|
beejaybee
Marvin
Reged: 18/07/2007
Posts: 4494
Loc: Really Here In Name Only
|
|
Quote:
If you're just talking about band-pass filters then the situation is surely just analogous to the current RGB filtration, except for the filters being centred on a different set of frequencies - where's the advantage? Or what am I missing?
OK. Suppose you have one photon each of red, green and blue light arriving at each photosite in a 2x2 Bayer matrix with RGB filters. Each photosite registers one photon - the one whose colour matches the filter.
Now, replace the red filter with cyan, the green filters with magenta and the blue filter with yellow. The photosite with the cyan filter registers both the green and blue photons; the magenta filtered photosites register both the red and blue photons, and the yellow filtered photosite registers both the red and green photons.
Total photons registered, 8 instead of 4.
|
huwevans
Old Hand
Reged: 05/08/2000
Posts: 15272
Loc: Dorset, UK
|
|
Only if they are band-stop filters - you're now talking about them being band-pass instead. Your magenta one at the very least must be a band-stop if its going to cut green whilst passing both red and blue.
But it's worse than that - even if you do make the yellow and cyan filters band-pass (and yes, in principle you could) then for them both to register the two nearest primary colours with significant response you'll be talking about a really large bandwidth, presumably with quite sharp roll-off, and that means you're going to have very poor ability to discriminate between frequencies within a pass-band. Possibly none at all.
So all in all it just isn't going to replicate the way that the human eye sees. Granted there are applications which don't require to do that, but for conventional photographic purposes we normally do, and the RGB array is designed essentially for that purpose.
--------------------
Huw Evans.
|
beejaybee
Marvin
Reged: 18/07/2007
Posts: 4494
Loc: Really Here In Name Only
|
|
Quote:
Your magenta one at the very least must be a band-stop if its going to cut green whilst passing both red and blue.
Not so, with interference filters you can have any shape response curve you wish. Including one which passes two bands with a blocked band between.
|
huwevans
Old Hand
Reged: 05/08/2000
Posts: 15272
Loc: Dorset, UK
|
|
Such a filter in this context is a band-stop filter - the point I'm making is unaffected by the distinction. All your filters necessarily have to pass a broad band including two primary colours, and that's the problem. Your pass-bands overlap with high values in the primary colours and that's what stops them being able to discriminate between different shades close to those primaries. A CMY sensor will tend to have poor discrimination where the human eye has good discrimination, and vice versa.
--------------------
Huw Evans.
|
beejaybee
Marvin
Reged: 18/07/2007
Posts: 4494
Loc: Really Here In Name Only
|
|
Quote:
Such a filter in this context is a band-stop filter
We're splitting hairs here.
A CMY filter set obviously needs to have a very carefully tailored response curve, otherwise adequate colour reproduction will be difficult to impossible to achieve. Actually the same applies to a RGB filter set, though a "good enough" response curve is much easier to achieve by use of dyes rather than interference. It's entirely a matter of whether the gain in light sensitivity is worth the relative difficulty of engineering an adequate CMY filter set.
|
huwevans
Old Hand
Reged: 05/08/2000
Posts: 15272
Loc: Dorset, UK
|
|
You're still missing the point - you're suggesting a filter system where for any given primary colour, two out of the three channels have low attentuation and the other one has high attenuation. That means that you're minimizing the difference in response in the only two channels that record that particular primary. But it is precisely that difference that enables us to differentiate between shades of that primary and indeed in between it and the next. Therefore a CMY filter array will have poor ability to discriminate colours in precisely those bands where the human eye is good.
There's really no way round this - what you're proposing (i) increases the bandwidth of the different filters, and (ii) makes the pass bands overlap more. That will get you more detected photons, for sure, but it will inevitably reduce the ability to discriminate between different frequencies.
--------------------
Huw Evans.
|
Fen
BAD WOLF
Reged: 12/03/2002
Posts: 20214
Loc: Currently Unknown!
|
|
Quote:
Quote:
Such a filter in this context is a band-stop filter
We're splitting hairs here.
Surely... that's splitting filters 
--------------------
Fen .......... My Galleries - My Blog - My Flickr
|
huwevans
Old Hand
Reged: 05/08/2000
Posts: 15272
Loc: Dorset, UK
|
|
Splitters!
--------------------
Huw Evans.
|
LargeFormat
old hand
Reged: 24/10/2006
Posts: 1066
Loc: Buckinghamshire and Cumbria
|
|
Quote:
New keyboard required...because my head has just exploded.
Always happens when Huw and BJB get going.
I went to an exhibition of autochromes a couple of weeks ago and there was an elegant demonstration of RGB using overlaid transparancies. I can do CMY separations which are normaly for additive printing. I'll print them on acetates to see how they compare with RGB.
|
huwevans
Old Hand
Reged: 05/08/2000
Posts: 15272
Loc: Dorset, UK
|
|
I notice Richard wanted a new keyboard, but considered a replacement or repaired head unnecessary. :-)
I rather like the response that has become traditional on Photo.net to threads with complicated arguments in them - the posting of the following picture:
:-)
--------------------
Huw Evans.
|
Nod
Carpal \'Tunnel
Reged: 08/04/2006
Posts: 4194
Loc: Devon, UK.
|
|
For posting here, a brane isn't always needed (but a SoH is a prime requisite!)
One common response to high flying threads designed to confuzzle less technical minded people on one forum I have been known to frequent is "Who gives a flying xxx?" especially when the current solution seems not to be a problem!
I'm sure that if the interference filter solution was practical, TV camera manufacturers would have adopted it IF there were any appreciable benefits.
--------------------
MATWSIJ.....
To avoid being offended, please insert apropriate smiley.
|
OneTen
'Two Breakfasts'
Reged: 23/06/2003
Posts: 2471
Loc: Devon
|
|
Quote:
I notice Richard wanted a new keyboard, but considered a replacement or repaired head unnecessary. :-)
Gaffer tape.
--------------------
Richard .......... My Website - My Blog - My Flickr
|
DaveS
enthusiast
Reged: 22/06/2007
Posts: 212
|
|
They already do, in the dichroic colour separation prisms used in all three-chip cameras (Ie, anything above micky-mouse level. However, the new breed of digital cinamatography cameras (Prices going from very nice car to very nice flat!) are starting to use single bayer-pattern sensors, though I have heard mutterings about Foveon type chips.
Dave
|
Previous
Index
Next
Threaded
Edit
Reply
Quote
