X3 How much benefit are you expecting? What am I missing?

[Peter G]

After reading all the recent "I am not buying a digicam until it is X3" followed by a bunch of "me toos". I think my expectations of this technology must be way off.

If I had to put number on it I would expect a 20% improvement from this technology. Worth putting purchases on hold for that?? For how long.

My expectations in more detail:

1: Monochromatic detail captured: No change.

Absolute B&W resolution is unlikely to be better. Current digital cameras are resolving things like power lines as thin as 1 pixel. You can't really improve on that.

2: Efficiency: No change.
This one I expect to be more contentious, so lets do a thought experiment:

Imagine 10 photons of each R,G,B ( some uniform gray shade) hitting each 10 micron cell. Lets look at 4 cells of each sensor.

All 4 Foveon X3 Cells see R10-G10-B10, Done, Perfect!

Now the mosaic only see one colour each. R10-G0-B0, etc... Now you average your Neighbors Green (10) and average your neighbors Blue (10).
So the end result is R10-G10-B10 after interpolation. Perfect. Same result.

3: Colour accuracy in non-boundary areas: No Change.

There is nothing in either design that leads to more accurate colour. That will probably stay within the quirks of the manufacturers preferences.

4: Colour detail. Nicely improved.

This is where mosaics have problems, when the average of your nieghbors is wrong. This will result in jagged Red-Green transitions, moire patterns. etc. X3 will solve this once and for all. Now that is great, but you have to ask just how much of an issue is this?

How bad is this problem that is being solved? My assesment is fairly minor, when it is there I would like it, But I won't be losing sleep because my next digital camear doesn't have X3.

I am very excited by the new technology and eagerly awaiting the Sigma SLR review, because I am a tech junkie, but I am not going to crawl into a hole for year or two waiting for all cameras to go X3 before emerging. Are you?

Peter

 

[PatiO]

After reading all the recent "I am not buying a digicam until it is
X3" followed by a bunch of "me toos". I think my expectations of
this technology must be way off.

If I had to put number on it I would expect a 20% improvement from
this technology. Worth putting purchases on hold for that?? For how
long.

My expectations in more detail:

1: Monochromatic detail captured: No change.
Absolute B&W resolution is unlikely to be better. Current digital
cameras are resolving things like power lines as thin as 1 pixel.
You can't really improve on that.

2: Efficiency: No change.
This one I expect to be more contentious, so lets do a thought
experiment:

Imagine 10 photons of each R,G,B ( some uniform gray shade) hitting
each 10 micron cell. Lets look at 4 cells of each sensor.

All 4 Foveon X3 Cells see R10-G10-B10, Done, Perfect!

Now the mosaic only see one colour each. R10-G0-B0, etc... Now you
average your Neighbors Green (10) and average your neighbors Blue
(10).
So the end result is R10-G10-B10 after interpolation. Perfect. Same
result.

3: Colour accuracy in non-boundary areas: No Change.
There is nothing in either design that leads to more accurate
colour. That will probably stay within the quirks of the
manufacturers preferences.

4: Colour detail. Nicely improved.
This is where mosaics have problems, when the average of your
nieghbors is wrong. This will result in jagged Red-Green
transitions, moire patterns. etc. X3 will solve this once and for
all. Now that is great, but you have to ask just how much of an
issue is this?

How bad is this problem that is being solved? My assesment is
fairly minor, when it is there I would like it, But I won't be
losing sleep because my next digital camear doesn't have X3.

I am very excited by the new technology and eagerly awaiting the
Sigma SLR review, because I am a tech junkie, but I am not going to
crawl into a hole for year or two waiting for all cameras to go X3
before emerging. Are you?

Nope! I'm going to lurk on eBay and pick up some bargains from those who do... :^) PatiO.

Peter

 

[Peter G]

Nope! I'm going to lurk on eBay and pick up some bargains from
those who do... :^) PatiO.

Excuse my obvious breakdown in logic. All of your cameras are obsolete, dump them while you can still get something for them! Hurry it will soon be too late!! ;-)

Peter

 

[Karl Guttag]

Peter,

First, I get the sense that logic is a lost cause in this forum.

I agree with you with the possible exception of your logic on #2. The issue to me is Signal to Noise. On a 4x4 region, the Foveon would see 4 x 10 "photons" of each of R, G, and B or 40 photons. The bayer system would see 20 Of Green and 10 of Red and 10 of Blue. Getting 4 stamples of each would let them average out the noise better so this would SEEM to be an advantage for the X3.

BUT, if we look at the Foveon Patent and consider the layout of a Pixel, the area that can capture light of a Foveon X3 pixel appears to be greatly reduced. First there is the issue of fitting 9 transitors into the cell. Then there is the issue of the concentric diffusions/wells that form the photo sensing diodes. The active light catching area of the top diode appears to be less than 1/4th of the pixel (of course this is 4 years old or so and I don't know how it scales -- this may be why they went to .18 micron ahead of others, to reduce the amount of light blockage by the trasistors).

When I net it all out there are probably enough unknown factors that it is hard to know.

The patent is at:

http://l2.espacenet.com/dips/bnsviewer?CY=ep&LG=en&DB=EPD&PN=US5965875&ID=US+++5965875A1+I+

A layout is on Page 8/16. The whole patent is interesting for those that want to undestand a bit about it. It has some nice charts and information. One thing that is interesting from the charts is how it looks like they will have to do a different kind of processing to separate the colors out (the "layers" don't appear to have a "sharp" frequency response).

I suspect they have played with doping the layers to enhance things since this patent was filed.

Karl

After reading all the recent "I am not buying a digicam until it is
X3" followed by a bunch of "me toos". I think my expectations of
this technology must be way off.

If I had to put number on it I would expect a 20% improvement from
this technology. Worth putting purchases on hold for that?? For how
long.

My expectations in more detail:

1: Monochromatic detail captured: No change.
Absolute B&W resolution is unlikely to be better. Current digital
cameras are resolving things like power lines as thin as 1 pixel.
You can't really improve on that.

2: Efficiency: No change.
This one I expect to be more contentious, so lets do a thought
experiment:

Imagine 10 photons of each R,G,B ( some uniform gray shade) hitting
each 10 micron cell. Lets look at 4 cells of each sensor.

All 4 Foveon X3 Cells see R10-G10-B10, Done, Perfect!

Now the mosaic only see one colour each. R10-G0-B0, etc... Now you
average your Neighbors Green (10) and average your neighbors Blue
(10).
So the end result is R10-G10-B10 after interpolation. Perfect. Same
result.

3: Colour accuracy in non-boundary areas: No Change.
There is nothing in either design that leads to more accurate
colour. That will probably stay within the quirks of the
manufacturers preferences.

4: Colour detail. Nicely improved.
This is where mosaics have problems, when the average of your
nieghbors is wrong. This will result in jagged Red-Green
transitions, moire patterns. etc. X3 will solve this once and for
all. Now that is great, but you have to ask just how much of an
issue is this?

How bad is this problem that is being solved? My assesment is
fairly minor, when it is there I would like it, But I won't be
losing sleep because my next digital camear doesn't have X3.

I am very excited by the new technology and eagerly awaiting the
Sigma SLR review, because I am a tech junkie, but I am not going to
crawl into a hole for year or two waiting for all cameras to go X3
before emerging. Are you?

Peter

--Karl

 

[jeff-c]

If you're comparing Foveon sensor to a full frame sensor yes it does look like the sensing area is smaller than the pixel but it's about 3/4 instead of 1/4 from what I see. However, if we're comparing a Foveon sensor to an interline sensor which is used in most consumer grade digital camera, its light catching area is bigger as interline sensor typically has almost half of the pixel area for storage transfer purpose.

BUT, if we look at the Foveon Patent and consider the layout of a
Pixel, the area that can capture light of a Foveon X3 pixel appears
to be greatly reduced. First there is the issue of fitting 9
transitors into the cell. Then there is the issue of the
concentric diffusions/wells that form the photo sensing diodes.
The active light catching area of the top diode appears to be less
than 1/4th of the pixel (of course this is 4 years old or so and I
don't know how it scales -- this may be why they went to .18 micron
ahead of others, to reduce the amount of light blockage by the
trasistors).> Karl

 

[Phil Askey]

My expectations in more detail:

1: Monochromatic detail captured: No change.
Absolute B&W resolution is unlikely to be better. Current digital
cameras are resolving things like power lines as thin as 1 pixel.
You can't really improve on that.

Not according to all our resolution tests, even the very best four megapixel digital cameras can't manage more than 75% of their published resolution. Because mosaic interpolation does affect monochrome resolution the Foveon chip SHOULD be better.

2: Efficiency: No change.

Probably, can't tell yet.

3: Colour accuracy in non-boundary areas: No Change.
There is nothing in either design that leads to more accurate
colour. That will probably stay within the quirks of the
manufacturers preferences.

Wrong, definite improvements. Consider this: for a pixel to detect blue it has to be under the blue part of the pixel array, two pixels side by side can not record continuous blue, thus some guesswork is made by the interpolation process. My own tests showed me that the X3 is better at continuous tone colour (stronger, cleaner).

4: Colour detail. Nicely improved.

Yes.

 

[Phil Askey]

How come nobody is talking about microlenses?

BUT, if we look at the Foveon Patent and consider the layout of a
Pixel, the area that can capture light of a Foveon X3 pixel appears
to be greatly reduced. First there is the issue of fitting 9
transitors into the cell. Then there is the issue of the
concentric diffusions/wells that form the photo sensing diodes.
The active light catching area of the top diode appears to be less
than 1/4th of the pixel (of course this is 4 years old or so and I
don't know how it scales -- this may be why they went to .18 micron
ahead of others, to reduce the amount of light blockage by the
trasistors).> Karl

 

[Peter G]

My expectations in more detail:

1: Monochromatic detail captured: No change.
Absolute B&W resolution is unlikely to be better. Current digital
cameras are resolving things like power lines as thin as 1 pixel.
You can't really improve on that.

Not according to all our resolution tests, even the very best four
megapixel digital cameras can't manage more than 75% of their
published resolution. Because mosaic interpolation does affect
monochrome resolution the Foveon chip SHOULD be better.

Ok, but how much of that is lens and how much is mosaic issues. Someone just recently published a D30 picture where they had power lines resolved down to about one pixel wide. There may be some improvement, I would not expect anything too dramatic here. Even if another 20% were recovered, it does not cover the fact that consumer sensor is 1.3MP.

2: Efficiency: No change.

Probably, can't tell yet.

I am hypothesizing. I don't see a clear cut advantage to either tech.

3: Colour accuracy in non-boundary areas: No Change.
There is nothing in either design that leads to more accurate
colour. That will probably stay within the quirks of the
manufacturers preferences.

Wrong, definite improvements. Consider this: for a pixel to detect
blue it has to be under the blue part of the pixel array, two
pixels side by side can not record continuous blue, thus some
guesswork is made by the interpolation process. My own tests
showed me that the X3 is better at continuous tone colour
(stronger, cleaner).

I can't argue with what you have witnessed. But it still could be the result of good colour algorithms. In a continuous tone situations the interpolation should still be correct. Its where there are variations that the mosaic should fall apart. Now if there is a gradient, again the X3 would be superior.

It is amazing tech, but I don't think we should all throw the mosaic cameras just yet. Especially since I don't expect a decent X3 3MP+ camera this year, but then you would know more about that than me.

Peter

 

[Karl Guttag]

How come nobody is talking about microlenses?

Because I foregot about them until I saw your post in the other thread.

Phil, you could save us a lot of time if you would just post a review comparing all the cameras being announced at PMA at various ISO settings and shooting situations I guess you are embargoed.

Karl

--Karl

 

[Peter G]

Karl,

In the end you simply get a number. It will unquestionably be more accurate to have all the readings rather than interpolate them. But if you interpolate by averaging all your neighbours to get a 12, or record an actual 12, the efficiency doesn't change. Now if that should have been some other number, you will introduce artifacts. I still maintain that either technology should produce the same results (colour and efficiency) on continuous tone areas. If all else(filtering efficiency/purity, light gathering area after factoring microlenses ) is equal.

Thanks, the patent was interesting. Doesn't Kodak also have a patent on the same idea? Could this have been found last year? I notice it is dated 1999.

There could be issues with this technology. They always point out the benefits, never the drawbacks.

I do look forward to the shake-up this may cause (and a lot of sweating over at Sony).

Peter

Peter,

First, I get the sense that logic is a lost cause in this forum.

I agree with you with the possible exception of your logic on #2.
The issue to me is Signal to Noise. On a 4x4 region, the Foveon
would see 4 x 10 "photons" of each of R, G, and B or 40 photons.
The bayer system would see 20 Of Green and 10 of Red and 10 of
Blue. Getting 4 stamples of each would let them average out the
noise better so this would SEEM to be an advantage for the X3.

BUT, if we look at the Foveon Patent and consider the layout of a
Pixel, the area that can capture light of a Foveon X3 pixel appears
to be greatly reduced. First there is the issue of fitting 9
transitors into the cell. Then there is the issue of the
concentric diffusions/wells that form the photo sensing diodes.
The active light catching area of the top diode appears to be less
than 1/4th of the pixel (of course this is 4 years old or so and I
don't know how it scales -- this may be why they went to .18 micron
ahead of others, to reduce the amount of light blockage by the
trasistors).

When I net it all out there are probably enough unknown factors
that it is hard to know.

The patent is at:

http://l2.espacenet.com/dips/bnsviewer?CY=ep&LG=en&DB=EPD&PN=US5965875&ID=US+++5965875A1+I+

A layout is on Page 8/16. The whole patent is interesting for
those that want to undestand a bit about it. It has some nice
charts and information. One thing that is interesting from the
charts is how it looks like they will have to do a different kind
of processing to separate the colors out (the "layers" don't appear
to have a "sharp" frequency response).
I suspect they have played with doping the layers to enhance things
since this patent was filed.

Karl

After reading all the recent "I am not buying a digicam until it is
X3" followed by a bunch of "me toos". I think my expectations of
this technology must be way off.

If I had to put number on it I would expect a 20% improvement from
this technology. Worth putting purchases on hold for that?? For how
long.

My expectations in more detail:

1: Monochromatic detail captured: No change.
Absolute B&W resolution is unlikely to be better. Current digital
cameras are resolving things like power lines as thin as 1 pixel.
You can't really improve on that.

2: Efficiency: No change.
This one I expect to be more contentious, so lets do a thought
experiment:

Imagine 10 photons of each R,G,B ( some uniform gray shade) hitting
each 10 micron cell. Lets look at 4 cells of each sensor.

All 4 Foveon X3 Cells see R10-G10-B10, Done, Perfect!

Now the mosaic only see one colour each. R10-G0-B0, etc... Now you
average your Neighbors Green (10) and average your neighbors Blue
(10).
So the end result is R10-G10-B10 after interpolation. Perfect. Same
result.

3: Colour accuracy in non-boundary areas: No Change.
There is nothing in either design that leads to more accurate
colour. That will probably stay within the quirks of the
manufacturers preferences.

4: Colour detail. Nicely improved.
This is where mosaics have problems, when the average of your
nieghbors is wrong. This will result in jagged Red-Green
transitions, moire patterns. etc. X3 will solve this once and for
all. Now that is great, but you have to ask just how much of an
issue is this?

How bad is this problem that is being solved? My assesment is
fairly minor, when it is there I would like it, But I won't be
losing sleep because my next digital camear doesn't have X3.

I am very excited by the new technology and eagerly awaiting the
Sigma SLR review, because I am a tech junkie, but I am not going to
crawl into a hole for year or two waiting for all cameras to go X3
before emerging. Are you?

Peter

--
Karl

 

[Homer40784]

Peter, I know I'm missing something here.

Karl,

In the end you simply get a number. It will unquestionably be more
accurate to have all the readings rather than interpolate them. But
if you interpolate by averaging all your neighbours to get a 12, or
record an actual 12, the efficiency doesn't change.

It seems to my poor technicians brain that it is FAR more "effecient" to just read a number than to calculate the average.

What am I missing?

Homer

 

[Peter iNova56624]

After reading all the recent "I am not buying a digicam until it is
X3" followed by a bunch of "me toos". I think my expectations of
this technology must be way off.

If I had to put number on it I would expect a 20% improvement from
this technology. Worth putting purchases on hold for that?? For how
long.

My expectations in more detail:

1: Monochromatic detail captured: No change.

Hmm. Are we looking at Phil's shots? Great tonality here.

Absolute B&W resolution is unlikely to be better. Current digital
cameras are resolving things like power lines as thin as 1 pixel.
You can't really improve on that.

2: Efficiency: No change.
This one I expect to be more contentious, so lets do a thought
experiment:

Imagine 10 photons of each R,G,B ( some uniform gray shade) hitting
each 10 micron cell. Lets look at 4 cells of each sensor.

All 4 Foveon X3 Cells see R10-G10-B10, Done, Perfect!

Now the mosaic only see one colour each. R10-G0-B0, etc... Now you
average your Neighbors Green (10) and average your neighbors Blue
(10).
So the end result is R10-G10-B10 after interpolation. Perfect. Same
result.

Except that the Bayer topped cells do a 60% green, 29% red and 11% blue computation to derive luminance, then separately do a number of color-within-the-luminance-lines computations to float color into it.

Meaning: It isn't as simple as you note, but the Fove sees each pixel as an unambiguous luminance and chroma source.

3: Colour accuracy in non-boundary areas: No Change.
There is nothing in either design that leads to more accurate
colour. That will probably stay within the quirks of the
manufacturers preferences.

If you stare at most sensitivity-challenged images in the shadow areas, you may see a blotchy color pattern unique to digicams. I don't see it in Phil's test shots.

4: Colour detail. Nicely improved.
This is where mosaics have problems, when the average of your
nieghbors is wrong. This will result in jagged Red-Green
transitions, moire patterns. etc. X3 will solve this once and for
all. Now that is great, but you have to ask just how much of an
issue is this?

The more you blow the image up, the bigger the issue becomes. But a 5-megapixel image at 5 X 7 print size will never see it. And images that are down-converted in the camera lose things like color noise, grain from high ISO and loose color detail. The Foveon images look like they were down-converted in Photoshop from much larger single chip images--but they are not.

How bad is this problem that is being solved? My assesment is
fairly minor, when it is there I would like it, But I won't be
losing sleep because my next digital camear doesn't have X3.

I think the Fove technology will become ubiquitous. Others may figure novel ways to make single-pixel, full-color sensors, so it isn't locked to Foveon, but it represents a new order of improved digi cam image. We all win.

-iNova

I am very excited by the new technology and eagerly awaiting the
Sigma SLR review, because I am a tech junkie, but I am not going to
crawl into a hole for year or two waiting for all cameras to go X3
before emerging. Are you?

Peter

 

[Terry Sessford]

It is amazing tech, but I don't think we should all throw the
mosaic cameras just yet. Especially since I don't expect a decent
X3 3MP+ camera this year, but then you would know more about that
than me.

Why will the Sigma SD-9 not be a "decent" camera? It's specs look good (assuming that you don't specialise in wide-angle photography due to 1.7x multiplier) and every review that I have seen of the SA-9 35mm SLR (on which the SD-9 is based) has been very good -- all say robust body, good handling, nice weight, logical control layout, good auto-focus and exposure, etc. And many of the Sigma lenses are very good.

Terry.

 

[Moshe Vainer]

Ha ha ha ha ha....,
Peter, of all the people, it's YOU that defends interpolation?

(for those who wonder what am i talking about look at Peter's messages in fuji forum)...

How about the sampling theory? how do you get 9 mega samples from 3 mega sensors without loose of detail?

Let me give you some examples:

We have a bayer sensor that looks like this:

GRGBGRGB
BGRGBGRG
GRGBGRGB
RGBGRGBG

And foveon sensor that looks like this:

PPPPPPPP
PPPPPPPP
PPPPPPPP
PPPPPPPP

(P is for pixel with RGB in it).

Now imagine that we have an image projected on a sensor with original data being

RGBGRGBG
GBGRGBGR
BGRGBGRG
GRGBGBGR

You have:
With Bayer sensor:

00000000
00000000
00000000
00000000

• A black area of 8x8 pixels

With foveon sensor:

RGBGRGBG
GBGRGBGR
BGRGBGRG
GRGBGBGR

• The exact image that was projected on the sensor

No advantage?

Now with your D30 example, imagine that this is not a single line,
but 3 lines 1 pixel wide, and they are red on a blue background:
So the original image is:

BBBBBBBB
RRRRRRRR
BBBBBBBB
RRRRRRRR
BBBBBBBB
RRRRRRRR

The foveon obviosly sees it as it is,
The Bayer however sees it as follows:

000B000B
00R000R0
000B000B
R000R000
000B000B
00R000R0

Now you tell me, should the interpolation algorithm produce:

BBBBBBBB
RRRRRRRR
BBBBBBBB
RRRRRRRR
BBBBBBBB
RRRRRRRR

Or may be

000B000B
00RB00RB
0R0B0R0B
R00BR00B
000B000B
00RB00RB

Or may be some other option?

Come, on, interpolation is good when there is no way not to interpolate, that's why SuperCCD is a bit better than regular CCD, but comparing interpolated with not interpolated is really funny, especially from you...

Rgrds,

Moshe

P.S., and see Phil's images, they really are stunning, whatever the theory is... (I just wish to know how they would be at iso800).

After reading all the recent "I am not buying a digicam until it is
X3" followed by a bunch of "me toos". I think my expectations of
this technology must be way off.

If I had to put number on it I would expect a 20% improvement from
this technology. Worth putting purchases on hold for that?? For how
long.

My expectations in more detail:

1: Monochromatic detail captured: No change.
Absolute B&W resolution is unlikely to be better. Current digital
cameras are resolving things like power lines as thin as 1 pixel.
You can't really improve on that.

2: Efficiency: No change.
This one I expect to be more contentious, so lets do a thought
experiment:

Imagine 10 photons of each R,G,B ( some uniform gray shade) hitting
each 10 micron cell. Lets look at 4 cells of each sensor.

All 4 Foveon X3 Cells see R10-G10-B10, Done, Perfect!

Now the mosaic only see one colour each. R10-G0-B0, etc... Now you
average your Neighbors Green (10) and average your neighbors Blue
(10).
So the end result is R10-G10-B10 after interpolation. Perfect. Same
result.

3: Colour accuracy in non-boundary areas: No Change.
There is nothing in either design that leads to more accurate
colour. That will probably stay within the quirks of the
manufacturers preferences.

4: Colour detail. Nicely improved.
This is where mosaics have problems, when the average of your
nieghbors is wrong. This will result in jagged Red-Green
transitions, moire patterns. etc. X3 will solve this once and for
all. Now that is great, but you have to ask just how much of an
issue is this?

How bad is this problem that is being solved? My assesment is
fairly minor, when it is there I would like it, But I won't be
losing sleep because my next digital camear doesn't have X3.

I am very excited by the new technology and eagerly awaiting the
Sigma SLR review, because I am a tech junkie, but I am not going to
crawl into a hole for year or two waiting for all cameras to go X3
before emerging. Are you?

Peter

 

[Peter G]

Sorry, I fogot to type "consumer" in front of camera. Not everyone will pay $3000 US for a camera.

It is amazing tech, but I don't think we should all throw the
mosaic cameras just yet. Especially since I don't expect a decent
X3 3MP+ camera this year, but then you would know more about that
than me.

Why will the Sigma SD-9 not be a "decent" camera? It's specs look
good (assuming that you don't specialise in wide-angle photography
due to 1.7x multiplier) and every review that I have seen of the
SA-9 35mm SLR (on which the SD-9 is based) has been very good --
all say robust body, good handling, nice weight, logical control
layout, good auto-focus and exposure, etc. And many of the Sigma
lenses are very good.

Terry.

 

[Peter G]

I wasn't thinking about computational efficiency, but light gathering efficiency. I really don't think that saving bayer interpolation computations will make that much difference on camera design.

Karl,

In the end you simply get a number. It will unquestionably be more
accurate to have all the readings rather than interpolate them. But
if you interpolate by averaging all your neighbours to get a 12, or
record an actual 12, the efficiency doesn't change.

It seems to my poor technicians brain that it is FAR more
"effecient" to just read a number than to calculate the average.

What am I missing?

Homer

 

[Peter G]

I think the issue is more that my base assumption is incorrect. In a true continuous tone situation there is no disadvantage to interpolation, but there really is no such thing in nature as pure continuous tone, it is all a series of subtle variations.

After reading all the recent "I am not buying a digicam until it is
X3" followed by a bunch of "me toos". I think my expectations of
this technology must be way off.

If I had to put number on it I would expect a 20% improvement from
this technology. Worth putting purchases on hold for that?? For how
long.

My expectations in more detail:

1: Monochromatic detail captured: No change.

Hmm. Are we looking at Phil's shots? Great tonality here.

Absolute B&W resolution is unlikely to be better. Current digital
cameras are resolving things like power lines as thin as 1 pixel.
You can't really improve on that.

2: Efficiency: No change.
This one I expect to be more contentious, so lets do a thought
experiment:

Imagine 10 photons of each R,G,B ( some uniform gray shade) hitting
each 10 micron cell. Lets look at 4 cells of each sensor.

All 4 Foveon X3 Cells see R10-G10-B10, Done, Perfect!

Now the mosaic only see one colour each. R10-G0-B0, etc... Now you
average your Neighbors Green (10) and average your neighbors Blue
(10).
So the end result is R10-G10-B10 after interpolation. Perfect. Same
result.

Except that the Bayer topped cells do a 60% green, 29% red and 11%
blue computation to derive luminance, then separately do a number
of color-within-the-luminance-lines computations to float color
into it.

Meaning: It isn't as simple as you note, but the Fove sees each
pixel as an unambiguous luminance and chroma source.

3: Colour accuracy in non-boundary areas: No Change.
There is nothing in either design that leads to more accurate
colour. That will probably stay within the quirks of the
manufacturers preferences.

If you stare at most sensitivity-challenged images in the shadow
areas, you may see a blotchy color pattern unique to digicams. I
don't see it in Phil's test shots.

4: Colour detail. Nicely improved.
This is where mosaics have problems, when the average of your
nieghbors is wrong. This will result in jagged Red-Green
transitions, moire patterns. etc. X3 will solve this once and for
all. Now that is great, but you have to ask just how much of an
issue is this?

The more you blow the image up, the bigger the issue becomes. But a
5-megapixel image at 5 X 7 print size will never see it. And images
that are down-converted in the camera lose things like color noise,
grain from high ISO and loose color detail. The Foveon images look
like they were down-converted in Photoshop from much larger single
chip images--but they are not.

How bad is this problem that is being solved? My assesment is
fairly minor, when it is there I would like it, But I won't be
losing sleep because my next digital camear doesn't have X3.

I think the Fove technology will become ubiquitous. Others may
figure novel ways to make single-pixel, full-color sensors, so it
isn't locked to Foveon, but it represents a new order of improved
digi cam image. We all win.

-iNova

I am very excited by the new technology and eagerly awaiting the
Sigma SLR review, because I am a tech junkie, but I am not going to
crawl into a hole for year or two waiting for all cameras to go X3
before emerging. Are you?

Peter

 

[Peter G]

Come on Moshe, if you completely read my message, you will note that I said where there is continuous tone (ie all the same) then interpolation will be as good as the real data. My last point is where I point out the obvious disadvantages of mosaic, which are significant.

I am trying to determine how large is the problem we are trying to solve. Doesn't warrant the wholesale junking of current digital cameras and stopping sales until all cameras are X3 as many seem to advocate. I just think people or over-reacting.

Comparing the images to those from a D30, they don't strike me as being that much better. Especially since we dont' know the conditions. The following are admittedly resized, but so are Foveons apparently:

http://photography-on-the.net/gallery/photo.php?photo=213&JS_SUBMIT_size_change=ok&exhibition=1&pass=public&size=quarter_hq

http://photography-on-the.net/gallery/photo.php?photo=210&JS_SUBMIT_size_change=ok&exhibition=1&pass=public&size=quarter

As far as defending interpolation, don't get me wrong, this technology is clearly better (with current info we have), but I think it is important to maintain a balanced view of the actual benefit.

How do you think a 3MP X3 will compare to a downsized 6MP d60 image?

Peter

How about the sampling theory? how do you get 9 mega samples from 3
mega sensors without loose of detail?

Let me give you some examples:

We have a bayer sensor that looks like this:

GRGBGRGB
BGRGBGRG
GRGBGRGB
RGBGRGBG

And foveon sensor that looks like this:

PPPPPPPP
PPPPPPPP
PPPPPPPP
PPPPPPPP

(P is for pixel with RGB in it).

Now imagine that we have an image projected on a sensor with
original data being

RGBGRGBG
GBGRGBGR
BGRGBGRG
GRGBGBGR

You have:
With Bayer sensor:

00000000
00000000
00000000
00000000

• A black area of 8x8 pixels

With foveon sensor:

RGBGRGBG
GBGRGBGR
BGRGBGRG
GRGBGBGR

• The exact image that was projected on the sensor

No advantage?

Now with your D30 example, imagine that this is not a single line,
but 3 lines 1 pixel wide, and they are red on a blue background:
So the original image is:

BBBBBBBB
RRRRRRRR
BBBBBBBB
RRRRRRRR
BBBBBBBB
RRRRRRRR

The foveon obviosly sees it as it is,
The Bayer however sees it as follows:

000B000B
00R000R0
000B000B
R000R000
000B000B
00R000R0

Now you tell me, should the interpolation algorithm produce:

BBBBBBBB
RRRRRRRR
BBBBBBBB
RRRRRRRR
BBBBBBBB
RRRRRRRR

Or may be

000B000B
00RB00RB
0R0B0R0B
R00BR00B
000B000B
00RB00RB

Or may be some other option?

Come, on, interpolation is good when there is no way not to
interpolate, that's why SuperCCD is a bit better than regular CCD,
but comparing interpolated with not interpolated is really funny,
especially from you...

Rgrds,

Moshe

P.S., and see Phil's images, they really are stunning, whatever the
theory is... (I just wish to know how they would be at iso800).

After reading all the recent "I am not buying a digicam until it is
X3" followed by a bunch of "me toos". I think my expectations of
this technology must be way off.

If I had to put number on it I would expect a 20% improvement from
this technology. Worth putting purchases on hold for that?? For how
long.

My expectations in more detail:

1: Monochromatic detail captured: No change.
Absolute B&W resolution is unlikely to be better. Current digital
cameras are resolving things like power lines as thin as 1 pixel.
You can't really improve on that.

2: Efficiency: No change.
This one I expect to be more contentious, so lets do a thought
experiment:

Imagine 10 photons of each R,G,B ( some uniform gray shade) hitting
each 10 micron cell. Lets look at 4 cells of each sensor.

All 4 Foveon X3 Cells see R10-G10-B10, Done, Perfect!

Now the mosaic only see one colour each. R10-G0-B0, etc... Now you
average your Neighbors Green (10) and average your neighbors Blue
(10).
So the end result is R10-G10-B10 after interpolation. Perfect. Same
result.

3: Colour accuracy in non-boundary areas: No Change.
There is nothing in either design that leads to more accurate
colour. That will probably stay within the quirks of the
manufacturers preferences.

4: Colour detail. Nicely improved.
This is where mosaics have problems, when the average of your
nieghbors is wrong. This will result in jagged Red-Green
transitions, moire patterns. etc. X3 will solve this once and for
all. Now that is great, but you have to ask just how much of an
issue is this?

How bad is this problem that is being solved? My assesment is
fairly minor, when it is there I would like it, But I won't be
losing sleep because my next digital camear doesn't have X3.

I am very excited by the new technology and eagerly awaiting the
Sigma SLR review, because I am a tech junkie, but I am not going to
crawl into a hole for year or two waiting for all cameras to go X3
before emerging. Are you?

Peter

 

[ianR]

I am not going to
crawl into a hole for year or two waiting for all cameras to go X3
before emerging. Are you?

No. It looks exciting but there are still so many unanswered questions. Can anybody predict if the new sensor is going to do its work fast for example? What if you get a great image but the processing time for all this is really slow?
Nothing is created that's perfect. What are going to be the drawbacks with this?
Reliability? Hot pixels? Low light?

The questions are endless and I think it's important to wait for some real answers.
Regards
Ian

 

[Moshe Vainer]

Peter, see my comments inside:

Come on Moshe, if you completely read my message, you will note
that I said where there is continuous tone (ie all the same) then
interpolation will be as good as the real data.

I have to disagree - If the continues tone is in the color of one of the sensors (especially red or blue), how the algorithm should know whether the tone is continues, or are there separate colored dots on a black background in an image? It's a guesswork. The algorithm may decide, that the most common case would be continues tone, and thus prefer this case to other cases. Just like fuji's SuperCCD would prefer horizontal/vertical lines over diagonal in it's interpolation. Statistically, images will improve (and that's why SuperCCD works), but you have no absolute guarantee, you win some, you loose some.

My last point is
where I point out the obvious disadvantages of mosaic, which are
significant.

Agreed.

I am trying to determine how large is the problem we are trying to
solve. Doesn't warrant the wholesale junking of current digital
cameras and stopping sales until all cameras are X3 as many seem to
advocate. I just think people or over-reacting.

People are excited with this development, and with the few samples we have seen so far. Of course they overreact.

Comparing the images to those from a D30, they don't strike me as
being that much better. Especially since we dont' know the
conditions. The following are admittedly resized, but so are
Foveons apparently:

http://photography-on-the.net/gallery/photo.php?photo=213&JS_SUBMIT_size_change=ok&exhibition=1&pass=public&size=quarter_hq

http://photography-on-the.net/gallery/photo.php?photo=210&JS_SUBMIT_size_change=ok&exhibition=1&pass=public&size=quarter

Well, your images are so much resized that this really is an indication of nothing. They are equivalent to taking every 4 pixels from GRGB to produce one real pixel, thus reducing resolution 4 times.

As for resized, look at the samples from Phil, those are real crops, and they are nothing short of amazing. I had a d30 for 2 weeks, and was able to make some shots with it. I really loved that camera, but it's quality is not on the same level as what we can see from (admittedly few) X3 samples.

On another note, D30 achieves it's smoothnes by loosing quite a bit of resolution, for 3mp sensor, it has very low resolving ability. Almost as if it was bayer interpolated, downsampled, and then upsampled back (I know it was not, but the antialias filter and bayer algorithm produce smoothness by loosing resolution). Here the case seems to be different. The foveon samples show BOTH detail and smoothness. Very formidable IMHO.

As far as defending interpolation, don't get me wrong, this
technology is clearly better (with current info we have), but I
think it is important to maintain a balanced view of the actual
benefit.

ok...

How do you think a 3MP X3 will compare to a downsized 6MP d60 image?

I don't know, we have not seen d60 images at all, if the camera even exists (it's all guesswork for now, ain't it?).

Assuming the reduction of individual sensors would not introduce more noise compared to d30, etc.. etc.., i still think you would need anything between 9 and 12 mp of bayer sensor to get the same quality as comparable foveon sensor.

Don't get me wrong, even 6mp sensor will probably resolve more detail than 3mp foveon, but to have clean image you would have to downsample bayer pattern, by binning together GRGB into a single pixel to have the same quality as foveon.

Even today, after foveon announcement, as much excited as i am about it, if at the same price, i would have to choose 16mp bayer or 3mp foveon, i'd certainly go for 16mp bayer and downsample it to around 4-5 mp, giving me quality equal to 4-5mp foveon (downsampling can be done simply by printing small enough on inkjet.) But 3mp foveon vs 6mp bayer... I don't know. Depends on how both deliver in the high iso department. So far foveon seems good (try to push the samples in photoshop, 1 stop brings almost no noise at all!!).

Rgrds,
Moshe