SuperCCD interpolation explained again

[ianR]

There have been lots of comments about Fuji Super CCD and that it’s just interpolation at 6mp, it’s not really 6mp and what interpolation means. Here’s my crack at it.

Imagine you’ve lined up a load of snooker balls into a square. We’ll give them numbers the same as on the snooker table:
Red=1, Yellow=2, Green=3, Brown=4, Blue =5, Pink=6, Black=7
Here’s our first square with a ball missing.
Blue is 5.

5,5,5,5,5
5,5, ? ,5,5
5,5,5,5,5

What colour is it do you think? Blue you say. Good, what you’ve just done is interpolation. You’ve calculated the missing colour from what you see in the colours around it.

Now have a look at this. This time we’ve just got a square of 4 colours around the one missing in the centre. I can tell you that the snooker balls reflect the colour of the missing ball and it changes their colour slightly. So if they are blue and the missing ball is brown, then they’ll be a slightly brownie blue.
4.75, 4.75
?
4.75, 4.75

The four balls are slightly less than 5 which is on the brown side of 5, (brown is 4, blue is 5) so they are in fact brownie blue.

Now you’ll see you’ve got 2 possible answers for what the ball in the centre is.

Your two answers are that the snooker ball in the middle is either a brownie blue the same as the rest of them, or that the ball in the middle is brown .
So, you can represent the two answers thus:
Answer 1 where all the balls are brownie blue
4.75, 4.75
4.75
4. 75, 4.75

Answer 2 where the snooker ball in the middle is actually brown and the others are brownie blue because of the reflections.

4.75, 4.75
4
4.75, 4.75

Now answer 1 is what interpolation does in Photoshop and elsewhere. It looks mathematically at the balls around the missing one and gives you the average of what it sees. We know this is wrong in this case because we know a snooker ball can actually only be blue or brown and nothing in between.

However what I believe the Fuji interpolation does is more like Answer 2. It looks at the snooker balls- or really the photosites- the way we do. It says, if there’s a bit of brown coming from that direction, and another bit of brown coming from that direction, then what’s in between must be brown.

Now you can see that what we get from Fuji interpolation is actually adding detail rather than simply adding information, and that 6mp resolution is really a higher resolution than 3.3mp. And more importantly, it's accurate.
I hope that helps
Ian

 

[Alias]

Quite an analogy Ian!!

Its a pity some of the knockers in the trade press didn'thave the chance to read this prior to putting fingers to keyboard.

Well done,

There have been lots of comments about Fuji Super CCD and that
it’s just interpolation at 6mp, it’s not really 6mp and
what interpolation means. Here’s my crack at it.
Imagine you’ve lined up a load of snooker balls into a
square. We’ll give them numbers the same as on the snooker
table:
Red=1, Yellow=2, Green=3, Brown=4, Blue =5, Pink=6, Black=7
Here’s our first square with a ball missing.
Blue is 5.

5,5,5,5,5
5,5, ? ,5,5
5,5,5,5,5
What colour is it do you think? Blue you say. Good, what
you’ve just done is interpolation. You’ve calculated
the missing colour from what you see in the colours around it.
Now have a look at this. This time we’ve just got a square of
4 colours around the one missing in the centre. I can tell you that
the snooker balls reflect the colour of the missing ball and it
changes their colour slightly. So if they are blue and the missing
ball is brown, then they’ll be a slightly brownie blue.
4.75, 4.75
?
4.75, 4.75
The four balls are slightly less than 5 which is on the brown
side of 5, (brown is 4, blue is 5) so they are in fact brownie blue.
Now you’ll see you’ve got 2 possible answers for what
the ball in the centre is.
Your two answers are that the snooker ball in the middle is either
a brownie blue the same as the rest of them, or that the ball in
the middle is brown .
So, you can represent the two answers thus:
Answer 1 where all the balls are brownie blue
4.75, 4.75
4.75
4. 75, 4.75
Answer 2 where the snooker ball in the middle is actually brown
and the others are brownie blue because of the reflections.

4.75, 4.75
4
4.75, 4.75

Now answer 1 is what interpolation does in Photoshop and elsewhere.
It looks mathematically at the balls around the missing one and
gives you the average of what it sees. We know this is wrong in
this case because we know a snooker ball can actually only be blue
or brown and nothing in between.
However what I believe the Fuji interpolation does is more like
Answer 2. It looks at the snooker balls- or really the photosites-
the way we do. It says, if there’s a bit of brown coming from
that direction, and another bit of brown coming from that
direction, then what’s in between must be brown.
Now you can see that what we get from Fuji interpolation is
actually adding detail rather than simply adding information, and
that 6mp resolution is really a higher resolution than 3.3mp. And
more importantly, it's accurate.
I hope that helps
Ian

--PhilB

 

[Richard Dunn]

Its a pity some of the knockers in the trade press didn'thave the
chance to read this prior to putting fingers to keyboard.

Well done,

There have been lots of comments about Fuji Super CCD and that
it’s just interpolation at 6mp, it’s not really 6mp and
what interpolation means. Here’s my crack at it.
Imagine you’ve lined up a load of snooker balls into a
square. We’ll give them numbers the same as on the snooker
table:
Red=1, Yellow=2, Green=3, Brown=4, Blue =5, Pink=6, Black=7
Here’s our first square with a ball missing.
Blue is 5.

5,5,5,5,5
5,5, ? ,5,5
5,5,5,5,5
What colour is it do you think? Blue you say. Good, what
you’ve just done is interpolation. You’ve calculated
the missing colour from what you see in the colours around it.
Now have a look at this. This time we’ve just got a square of
4 colours around the one missing in the centre. I can tell you that
the snooker balls reflect the colour of the missing ball and it
changes their colour slightly. So if they are blue and the missing
ball is brown, then they’ll be a slightly brownie blue.
4.75, 4.75
?
4.75, 4.75
The four balls are slightly less than 5 which is on the brown
side of 5, (brown is 4, blue is 5) so they are in fact brownie blue.
Now you’ll see you’ve got 2 possible answers for what
the ball in the centre is.
Your two answers are that the snooker ball in the middle is either
a brownie blue the same as the rest of them, or that the ball in
the middle is brown .
So, you can represent the two answers thus:
Answer 1 where all the balls are brownie blue
4.75, 4.75
4.75
4. 75, 4.75
Answer 2 where the snooker ball in the middle is actually brown
and the others are brownie blue because of the reflections.

4.75, 4.75
4
4.75, 4.75

Now answer 1 is what interpolation does in Photoshop and elsewhere.
It looks mathematically at the balls around the missing one and
gives you the average of what it sees. We know this is wrong in
this case because we know a snooker ball can actually only be blue
or brown and nothing in between.
However what I believe the Fuji interpolation does is more like
Answer 2. It looks at the snooker balls- or really the photosites-
the way we do. It says, if there’s a bit of brown coming from
that direction, and another bit of brown coming from that
direction, then what’s in between must be brown.
Now you can see that what we get from Fuji interpolation is
actually adding detail rather than simply adding information, and
that 6mp resolution is really a higher resolution than 3.3mp. And
more importantly, it's accurate.
I hope that helps
Ian

--
PhilB

Good explanation Ian. I know Fuji inrepolation picks up detail which was there but was not visible at the lower resolution. Yes it sometimes gets it wrong especialy if light is low and this is why shadow detail is not brilliant but if the light is good I have not seen better more detailed unpixelated pictures from any camera less tha £1500.--Richard Dunn

 

[Alias]

Richard,

Interesting you pick up on the shadow detail problem.

I'm beginning to think this is down to the operator not being able to judge exactly which white balance and photometry setting is best to use in certain shadow circumstances, rather than the inability of the camera to cope - if it was set properly.

The simplistic answer may be that the camera should be set in auto mode but those of us who have tried to explore the operating envelope to the full know that Auto itself can be very simplistic in its interpretation.

I don't know the answer, other than experimentation but, for the most photogenic shots, how do you expriment before the moment has gone?

With film, you get to know how a film type will react with your SLR. Perhaps we haven't learnt the wrinkles of operating the super CCD as it should be operated yet and practice will make perfect, but I think the camera is probably a little too clever for us in offering so many white balance options and a choice of photometry without offering its own "best" settings to be accepted or rejected, on half depressing the shutter.

Perhaps auto bracketing is the logical way round, but its a long winded way of getting it right!

Good explanation Ian. I know Fuji inrepolation picks up detail
which was there but was not visible at the lower resolution. Yes it
sometimes gets it wrong especialy if light is low and this is why
shadow detail is not brilliant but if the light is good I have not
seen better more detailed unpixelated pictures from any camera less
tha £1500.
--
Richard Dunn

--PhilB

 

[Doug Jones]

Good analogy Ian. I too have thought about the Super CCD and I believe it is better because it has better symetry than the Bayer pattern - as follows
2 possible Bayer patterns are:

B G R G B G R G B G R G B . . . . . B G R G B G R G B G R G B
G B G R G B G R G B G R G . . . . . G B G R G B G R G B G R G
B G R G B G R G B G R G B . . . . . R G B G R G B G R G B G R

And the Super CCD staggered layout
(doing my best with the proportional font and HTML interference)

B . R . B . R . B . R . B . R . B . R . B . R . B
. G . G . G . G .G . G . G .G . G . G . G .G
R . B . R . B . R . B . R . B . R . B . R . B . R

In these examples if you just look at the immediate neighbours of any of the Green pixels you will find:

In the Super CCD every green has the other colours symetrically balanced around it.

In the Bayer pattern No 1 each Green has one and three of the other colours arount it - alternating in the pattern.

In the Bayer pattern No 2 there is an even number of each other colour but they are biased on either side of the Green.

I think this is why the Super CCD has such good continuous colour tone and why my older Coolpix 990 had such messy continuous tones.

If you extend this into interpolating "extra" pixels the geometry of the Super CCD looks pretty good. I haven't been able to mathematically figure it out yet but you will notice that the increase in pixels is about 2 which means the square root of 2 in each X and Y direction. And we know that a right angle triangle with adjacent sides of length 1 has a hypotenuse of root 2. The pixels on the Super CCD are effectively in a 45 degree tilted pattern so "regularising" it to a standard grid probably gives this root 2 squared (or doubling) of the number of pixels.

An important thing to remember is that to get the lower resolution 3M pictures requires another - probably equally as complicated - interpolation method. It is NOT the native pattern of the Super CCD. This is why I believe that the 6M image is just as valid as the 3M image and they are both equally interpolated.

Another important thing to remember is that so called "standard" CCDs can only produce images by interpolation anyway. They are "inventing" coloured pixels by interpolating from primary colour pixels. And I believe the Super CCD method and layout is far superior to the Bayer method.

My 2 bits worth.

Doug Jones
Canberra
http://www.panamagic.com.au

There have been lots of comments about Fuji Super CCD and that
it’s just interpolation at 6mp, it’s not really 6mp and
what interpolation means. Here’s my crack at it.
Imagine you’ve lined up a load of snooker balls into a
square. We’ll give them numbers the same as on the snooker
table:
Red=1, Yellow=2, Green=3, Brown=4, Blue =5, Pink=6, Black=7
Here’s our first square with a ball missing.
Blue is 5.

5,5,5,5,5
5,5, ? ,5,5
5,5,5,5,5
What colour is it do you think? Blue you say. Good, what
you’ve just done is interpolation. You’ve calculated
the missing colour from what you see in the colours around it.
Now have a look at this. This time we’ve just got a square of
4 colours around the one missing in the centre. I can tell you that
the snooker balls reflect the colour of the missing ball and it
changes their colour slightly. So if they are blue and the missing
ball is brown, then they’ll be a slightly brownie blue.
4.75, 4.75
?
4.75, 4.75
The four balls are slightly less than 5 which is on the brown
side of 5, (brown is 4, blue is 5) so they are in fact brownie blue.
Now you’ll see you’ve got 2 possible answers for what
the ball in the centre is.
Your two answers are that the snooker ball in the middle is either
a brownie blue the same as the rest of them, or that the ball in
the middle is brown .
So, you can represent the two answers thus:
Answer 1 where all the balls are brownie blue
4.75, 4.75
4.75
4. 75, 4.75
Answer 2 where the snooker ball in the middle is actually brown
and the others are brownie blue because of the reflections.

4.75, 4.75
4
4.75, 4.75

Now answer 1 is what interpolation does in Photoshop and elsewhere.
It looks mathematically at the balls around the missing one and
gives you the average of what it sees. We know this is wrong in
this case because we know a snooker ball can actually only be blue
or brown and nothing in between.
However what I believe the Fuji interpolation does is more like
Answer 2. It looks at the snooker balls- or really the photosites-
the way we do. It says, if there’s a bit of brown coming from
that direction, and another bit of brown coming from that
direction, then what’s in between must be brown.
Now you can see that what we get from Fuji interpolation is
actually adding detail rather than simply adding information, and
that 6mp resolution is really a higher resolution than 3.3mp. And
more importantly, it's accurate.
I hope that helps
Ian

--Doug JonesCanberra

 

[SFJP]

You are perfectly right. A standard CCD uses Bayer interpolation to decide of the complete color of each pixel (R-G-B value), as its photosites record only one color. The arrangement of the color grid on the Fuji superCCD helps in deciding more accurately of the complete RGB color of each pixel. Because the photosites of the superCCD are not aligned on a square gid but follow a honeycomb pattern, to get a square grid of pixels it is moreover neccessary to deduce new intermediary pixels. This is why the 6 megapixels image is the native image extracted from the 3.3 megaphotosites superCCD of the 6900 or S1.

However, I suspect that the 3 mp image is obtained just by reduction of the native 6 mp image, not by an alternative interpolation using directly the 3 mphotosites initial record, which means it necessarily loses information with respect to the 6 mp native one. In my view we have the following process:

3.3 mp honeycomb record ---> (extraction) 6 mp square grid image ---> (reduction 1/2) 3 mp image

Another point of importance in the honeycomb pattern and the orientation of the photosites is that each photosite is larger on the superCCD than it is on a conventional CCD of same area and same number of photosites. This plays an important role in signal/noise ratio. Moreover, the photosites are closer, which allows more accuracy in the "deduction/interpolation" of the missing pixels.

Jean-Paul

Good analogy Ian. I too have thought about the Super CCD and I
believe it is better because it has better symetry than the Bayer
pattern - as follows
2 possible Bayer patterns are:

B G R G B G R G B G R G B . . . . . B G R G B G R G B G R G B
G B G R G B G R G B G R G . . . . . G B G R G B G R G B G R G
B G R G B G R G B G R G B . . . . . R G B G R G B G R G B G R

And the Super CCD staggered layout
(doing my best with the proportional font and HTML interference)

B . R . B . R . B . R . B . R . B . R . B . R . B
. G . G . G . G .G . G . G .G . G . G . G .G
R . B . R . B . R . B . R . B . R . B . R . B . R

In these examples if you just look at the immediate neighbours of
any of the Green pixels you will find:
In the Super CCD every green has the other colours symetrically
balanced around it.
In the Bayer pattern No 1 each Green has one and three of the other
colours arount it - alternating in the pattern.
In the Bayer pattern No 2 there is an even number of each other
colour but they are biased on either side of the Green.

I think this is why the Super CCD has such good continuous colour
tone and why my older Coolpix 990 had such messy continuous tones.

If you extend this into interpolating "extra" pixels the geometry
of the Super CCD looks pretty good. I haven't been able to
mathematically figure it out yet but you will notice that the
increase in pixels is about 2 which means the square root of 2 in
each X and Y direction. And we know that a right angle triangle
with adjacent sides of length 1 has a hypotenuse of root 2. The
pixels on the Super CCD are effectively in a 45 degree tilted
pattern so "regularising" it to a standard grid probably gives this
root 2 squared (or doubling) of the number of pixels.

An important thing to remember is that to get the lower resolution
3M pictures requires another - probably equally as complicated -
interpolation method. It is NOT the native pattern of the Super
CCD. This is why I believe that the 6M image is just as valid as
the 3M image and they are both equally interpolated.

Another important thing to remember is that so called "standard"
CCDs can only produce images by interpolation anyway. They are
"inventing" coloured pixels by interpolating from primary colour
pixels. And I believe the Super CCD method and layout is far
superior to the Bayer method.

My 2 bits worth.

Doug Jones
Canberra
http://www.panamagic.com.au

 

[Redbreast [UK]]

However, I suspect that the 3 mp image is obtained just by
reduction of the native 6 mp image, not by an alternative
interpolation using directly the 3 mphotosites initial record,
which means it necessarily loses information with respect to the 6
mp native one. In my view we have the following process:

3.3 mp honeycomb record ---> (extraction) 6 mp square grid image
---> (reduction 1/2) 3 mp image

Another point of importance in the honeycomb pattern and the
orientation of the photosites is that each photosite is larger on
the superCCD than it is on a conventional CCD of same area and same
number of photosites. This plays an important role in signal/noise
ratio. Moreover, the photosites are closer, which allows more
accuracy in the "deduction/interpolation" of the missing pixels.

This has certainly been my view all along, which is why I had not agreed with initial posts (shortly after the 6900 was introduced) that the 3MP setting was the "baseline" for the SuperCCD.

It is also why when I contribute to queries about the "best" setting to use for one's standard, I have always maintained that the 6MP Fine is the optimum for this camera, since a) it is the closest to the camera's starting point and requires least jiggery-pokery by the on-board chips, and b) it derives maximum image-quality benefit from the unique technology of the SuperCCD (and for which we have paid good money!!)

I agree that there are circumstances where a lower MP setting is indicated, for instance where the 6MP's long delay in recording the image on the SM is unsuited to a fast changing scenario; or where available SM space is the key constraint; and so on.

But the standard setting should always be 6MP Fine, and all others treated as exceptions and deployed as such.

My compliments to Ian, Phil, Richard, Doug and Jean-Paul for bringing this out into the open in such a clear manner, well done chaps.--Regards,Robin [Redbreast]

 

[Richard Dunn]

Interesting you pick up on the shadow detail problem.

I'm beginning to think this is down to the operator not being able
to judge exactly which white balance and photometry setting is best
to use in certain shadow circumstances, rather than the inability
of the camera to cope - if it was set properly.

The simplistic answer may be that the camera should be set in auto
mode but those of us who have tried to explore the operating
envelope to the full know that Auto itself can be very simplistic
in its interpretation.

I don't know the answer, other than experimentation but, for the
most photogenic shots, how do you expriment before the moment has
gone?

With film, you get to know how a film type will react with your
SLR. Perhaps we haven't learnt the wrinkles of operating the super
CCD as it should be operated yet and practice will make perfect,
but I think the camera is probably a little too clever for us in
offering so many white balance options and a choice of photometry
without offering its own "best" settings to be accepted or
rejected, on half depressing the shutter.

Perhaps auto bracketing is the logical way round, but its a long
winded way of getting it right!

Good explanation Ian. I know Fuji inrepolation picks up detail
which was there but was not visible at the lower resolution. Yes it
sometimes gets it wrong especialy if light is low and this is why
shadow detail is not brilliant but if the light is good I have not
seen better more detailed unpixelated pictures from any camera less
tha £1500.
--
Richard Dunn

--
PhilB

Phil, I'm afraid I rarely experiment when taking a shot. I never change white balance and rarely bracket. I tend to focus on the composition and the image and try and hold the camera still. I do take several shots and sometimes use spot metering with different focal points and see later which is best. I will try altering colour balance now you mention it. Have you any other ideas how shadow detail can be improved apart from the use of more secondary lighting?
--Richard Dunn

 

[Tim O]

I've also thought that the necessity of having to interpolate from the honeycomb pattern into a rectangular pattern has the advantage that it is an intelligent upsampling that, hopefully, is computed using the (perhaps) 9 or 10 bits of data from the CCD sites before the image is truncated to the 8 bits maximum presented in the TIFF or JPEG format.

A "chap" from the west side of the pond.

 

[Alias]

Hi Richard,

This one can still be a bit of a lottery for me. If there is a very sharp contrast between the "in light" and "in shadow" areas of the shot (not often in Ireland!!) and if I have the time, I set the white balance for bright sunlight (if outside) or whatever type of light it will be (indoors) and take a reading from the bright area using spot metering.

I then set the white balance to overcast and take a spot metering reading from the shadow.

Depending on how sharp the contrast is I either average the result or lean towards more or less light, again depending on the result I want, and set either the shutter or the aperture accordingly.

It does work most of the time but you have to be taking landscapes or still life (or a very patient model) to make this work.

A faster method, but less precise, is to just average the result of a reading from both areas, without changing the white balance.

It really does come down to experimentation - but at least we don't have to buy film or pay for processing to see results!!

Interesting you pick up on the shadow detail problem.

I'm beginning to think this is down to the operator not being able
to judge exactly which white balance and photometry setting is best
to use in certain shadow circumstances, rather than the inability
of the camera to cope - if it was set properly.

The simplistic answer may be that the camera should be set in auto
mode but those of us who have tried to explore the operating
envelope to the full know that Auto itself can be very simplistic
in its interpretation.

I don't know the answer, other than experimentation but, for the
most photogenic shots, how do you expriment before the moment has
gone?

With film, you get to know how a film type will react with your
SLR. Perhaps we haven't learnt the wrinkles of operating the super
CCD as it should be operated yet and practice will make perfect,
but I think the camera is probably a little too clever for us in
offering so many white balance options and a choice of photometry
without offering its own "best" settings to be accepted or
rejected, on half depressing the shutter.

Perhaps auto bracketing is the logical way round, but its a long
winded way of getting it right!

Good explanation Ian. I know Fuji inrepolation picks up detail
which was there but was not visible at the lower resolution. Yes it
sometimes gets it wrong especialy if light is low and this is why
shadow detail is not brilliant but if the light is good I have not
seen better more detailed unpixelated pictures from any camera less
tha £1500.
--
Richard Dunn

--
PhilB

Phil, I'm afraid I rarely experiment when taking a shot. I never
change white balance and rarely bracket. I tend to focus on the
composition and the image and try and hold the camera still. I do
take several shots and sometimes use spot metering with different
focal points and see later which is best. I will try altering
colour balance now you mention it. Have you any other ideas how
shadow detail can be improved apart from the use of more secondary
lighting?

--
Richard Dunn

--PhilB

 

[ianR]

Thanks everyone for your positive feedback and for all the further work you've added to this. Let's hope we convince some of the doubters.

Tim, according to the Fuji specs, the CCD is 24 bit, although someone said recently that it was 12. Anyone know for sure.

Also, although we all call it a 'honeycomb' because that's what Fuji marketing label it as. It's actually an octagon not a hexagon. Does that make any difference to the rectangular grid maths JP?

And yes, converting from B&W to colour is, 'interpretation' if not actually 'interpolation' so the others needn't get so high and mighty about their pixel count. On top of that, FUJI COLOUR IS BEST!!!
Ian

I've also thought that the necessity of having to interpolate
from the honeycomb pattern into a rectangular pattern has the
advantage that it is an intelligent upsampling that, hopefully, is
computed using the (perhaps) 9 or 10 bits of data from the CCD
sites before the image is truncated to the 8 bits maximum
presented in the TIFF or JPEG format.

A "chap" from the west side of the pond.

 

[liza wallis]

Ian Ian Ian...

You are either a lunatic or a bloody genius. I am thinking perhaps a little of both. Your explanation is brilliant. I was particularly gripped when you got to the part about the "brownie blue." What I want to know now is how long you actually spent thinking that one up. So quaint it was. No matter how you slice it, this was the best dang interpretation of Fuji Interpolation I have ever read. It would further explain why this camera is so darn good.

I am having real trouble getting on to the forums here the last couple of days which explains why it is only now that I am getting to read this and am seeming perhaps a bit slow witted in my response.

What I want to now know is whether in fact you and JP are planning on linking up when he comes to live in France after the new year. If indeed you are then please alert the French government because I do think that the two of you together could prove to be a menace to society. First of all you are both just too darn smart, and second, you are both just way out of control!

Best regards

liza

PS... by the way, thank you for all the work you put into your explanation
--www.lizawallis.comwww.lizawallis.com/69er

 

[Lance A]

Would this info apply also to the 4700 and 4800? I believe a few folks have mentioned on this forum that the 4700/4800 SuperCCD differs from the 6800/6900.

Lance

 

[ianR]

Yes Lance it does apply.

These cameras have smaller resolution but they have exactly the same SuperCCD technology. The light falling between the photosites can be read and interpreted by more than one photosite making for a more accurate reconstitution of what's going on in between them.

Also, the wizzardry is taking place before .jpeg compression takes place, so you should get less loss than with ordinary interpolation afterwards.
regards
Ian

Would this info apply also to the 4700 and 4800? I believe a few
folks have mentioned on this forum that the 4700/4800 SuperCCD
differs from the 6800/6900.

Lance

 

[NickL]

However what I believe the Fuji interpolation does is more like
Answer 2. It looks at the snooker balls- or really the photosites-
the way we do. It says, if there’s a bit of brown coming from
that direction, and another bit of brown coming from that
direction, then what’s in between must be brown.

Ian,

I think I'm being slow; it's the last Friday before Christmas, and I'm looking forward to going home, but I'm not sure I understand the logic of your analogy. This is probably my failing, as I can still taste toothpaste in my mouth, which tells me it's too early...

However

How does the SuperCCD "realise" that the brownie blue photosites should actually be blue and are brownie blue through being tinged by the missing brown in the middle? I can see how you can make this assumption in a limited colourspace, but in a continuously variable colourspace, how could the camera "know" that it was a bit of brown that was reflecting on the blue sites, rather than the original colour was actually a brownie blue?

What I am asking is, "what happens if the original colour really is 4.75". How does the camera know which way to go?

I hope you can understand where I'm coming from. I'm not sure that I'm explaining myself well.

In short, I do not understand how the SuperCCD can make any intelligent guesses regarding the interpolated pixel in the middle in a continuously variable colourspace.

I can guess that the argument of reduction from 24 bit to 16 bit per channel colour is analagous to mapping into a reduced colourspace, but assuming the interpolation is performed at the original, full colourspace, how can any "intelligent" decision be made?

Perhaps I am taking the analogy too far, but snooker balls are reflective. What happens in a flat environment, avoiding specular reflection and ignoring the comparatively tiny effects of diffuse reflection?

I hope you understand where I'm coming from, and can put me right as I have to believe something better than mere average interpolation is happening, but want to know how

Cheers,

Nick.

 

[ianR]

Well I did make the analogy quite simple and I was expecting people to shout at me that the brown ball would aso be reflecting too, so what you say isn't really surprising.

I think the answer is that the brown bit is only on one side. The brown bit is just the reflection on one side of the sensor, and of course it wouldn't really produce a number like I gave. You've got 8 sides on a SuperCCD and I don't know exactly how it sees the world but I'll try to take my analagy one step further.
This is my original analagy
4.75, 4.75
4
4.75, 4.75
Now if you place a ball of 4.75 in the centre what you actually get is:
4.9375, 4.9375
4.75
4.9375, 4.9375

Remember we know the balls in the corners are actually 5. If they were really 4.75 then the ball in the middle really would have to be 4.75. Even if I've got the maths wrong which is entirely possible, you can't get the same result from two different colours as long as the sensors can 'communicate' or share information. In this case the extra bit of blue in the central ball takes away less of the blue in the balls around it than a purely brown ball.

In reality, this is happening all the time. Colours do interreact to give us a different perception of what they look like.

I have no idea if all this is true or not, it's just my theory based on what I've read and it's up to you to like it or not.
Thanks for all the comments
All the best
Ian

 

[Richard Dunn]

This one can still be a bit of a lottery for me. If there is a
very sharp contrast between the "in light" and "in shadow" areas of
the shot (not often in Ireland!!) and if I have the time, I set the
white balance for bright sunlight (if outside) or whatever type of
light it will be (indoors) and take a reading from the bright area
using spot metering.

I then set the white balance to overcast and take a spot metering
reading from the shadow.

Depending on how sharp the contrast is I either average the result
or lean towards more or less light, again depending on the result I
want, and set either the shutter or the aperture accordingly.

Thanks Phil, it sounds like it's as difficult for you as me. My problem is that I forget to reset any settings I change and finish up with a lot of pictures messed up. I After turning my ISO to 400 to take natural light shots in Siena's duopmo I forgot to change back and took about 50 shots in Rome in brilliant sunlight at ISO 400 before I realised. They weren't bad but could have been brilliant.
It does work most of the time but you have to be taking landscapes
or still life (or a very patient model) to make this work.

A faster method, but less precise, is to just average the result of
a reading from both areas, without changing the white balance.

It really does come down to experimentation - but at least we don't
have to buy film or pay for processing to see results!!

Interesting you pick up on the shadow detail problem.

I'm beginning to think this is down to the operator not being able
to judge exactly which white balance and photometry setting is best
to use in certain shadow circumstances, rather than the inability
of the camera to cope - if it was set properly.

The simplistic answer may be that the camera should be set in auto
mode but those of us who have tried to explore the operating
envelope to the full know that Auto itself can be very simplistic
in its interpretation.

I don't know the answer, other than experimentation but, for the
most photogenic shots, how do you expriment before the moment has
gone?

With film, you get to know how a film type will react with your
SLR. Perhaps we haven't learnt the wrinkles of operating the super
CCD as it should be operated yet and practice will make perfect,
but I think the camera is probably a little too clever for us in
offering so many white balance options and a choice of photometry
without offering its own "best" settings to be accepted or
rejected, on half depressing the shutter.

Perhaps auto bracketing is the logical way round, but its a long
winded way of getting it right!

Good explanation Ian. I know Fuji inrepolation picks up detail
which was there but was not visible at the lower resolution. Yes it
sometimes gets it wrong especialy if light is low and this is why
shadow detail is not brilliant but if the light is good I have not
seen better more detailed unpixelated pictures from any camera less
tha £1500.
--
Richard Dunn

--
PhilB

Phil, I'm afraid I rarely experiment when taking a shot. I never
change white balance and rarely bracket. I tend to focus on the
composition and the image and try and hold the camera still. I do
take several shots and sometimes use spot metering with different
focal points and see later which is best. I will try altering
colour balance now you mention it. Have you any other ideas how
shadow detail can be improved apart from the use of more secondary
lighting?

--
Richard Dunn

--
PhilB

--Richard Dunn

 

[Richard Dunn]

Would this info apply also to the 4700 and 4800? I believe a few
folks have mentioned on this forum that the 4700/4800 SuperCCD
differs from the 6800/6900.

Lance

Lance I think largely it does as the super ccd is a similart patter, However the ccd elements are larger in the 4700 though the signal processing is inferior. This causes IMO better colour but more noise and explains the 200 base ISO of the 4700.
--Richard Dunn