My own sensor idea

Another Sensor idea?

It occurs to me that the current sensors filter away more light than they really have to. Instead of the Bayer pattern RGB filter, how about this alternate filter:

W Y W Y W Y
C W C W C W
W Y W Y W Y
C W C W C W
W Y W Y W Y
C W C W C W
W Y W Y W Y
C W C W C W

The filters are white (no filter), yellow (lets in green and red), and cyan (lets in green and blue). This is "demosaiced" by the formula:

Red = W-C
Green = C+Y-W
Blue = W-Y

Where W, Y, and C are the white, yellow, and cyan sensor values. For each pixel, one of these will be a straight CCD value while the others will be averages of the 4 neighbors.

Luminance is R+G+B = (W-C)+(C+Y-W)+(W-Y) = W. This essentially means that maximum resolution is given to luminance, while chroma has half its resolution. This is what jpeg compression does first anyway, and for the good reason that human vision is more sensitive to luminance resolution than chroma resolution.

The resolution qualities are essentially similar to Bayer pattern RGB filtering.

The advantage, I think, is that it will capture more light than a normal RGB filter. This should translate to better sensitivity and lower noise, right?

Am I missing something? Is there some reason why traditional RGB filters are superior to this?
--
Isaac Kuo

Without thinking this through real thoroughly, I would suggest two things that might ruin it (though i might very well be wrong here).

The advantage of lower noise seems to diminish or even negate when you form any of the RGB by subtracting a filtered color (or adding two and subbing white). While white should have SQRT(3) lower noise than does either C or Y, the formation of R or B through subtraction of C or Y adds, I think, SQRT(2) noise. And in the formation of G, you break even. so you might have only a very very slim advantage in noise of SQRT(3/2) for the R & B channels.

The other problem might be in the lack of green bias that is normally incurred in a RGGB or CYGM type filter layering. But maybe you can acct for this in the addition/interpolation?

IsaacKuo said:

Another Sensor idea?

It occurs to me that the current sensors filter away more light
than they really have to. Instead of the Bayer pattern RGB filter,
how about this alternate filter:

Click to expand...

--
Photos, tips and tests at:
http://www.geocities.com/glowluzid

I like that some people are thinking about this. It's very cool.

Another issue--how does subtracting the colors, which have spectral responses, add to color artifacts? They're not just boxcar functions. They're not even gaussian spectral responses. But this might not be that big of an issue, depending on spectral overlap, and distribution functions...

The JVC digital video camcorders have a similar system, they call it "progressive color filter." They use Magenta, Cyan, Yellow, and White for the same reason, I suspect -- to get more light sensitivity.

see
http://www.jvc.com/product.jsp?modelId=MODL026045&page=2
for example on model GR-DV2000U.

-bruce

IsaacKuo said:

Another Sensor idea?

It occurs to me that the current sensors filter away more light
than they really have to. Instead of the Bayer pattern RGB filter,
how about this alternate filter:

W Y W Y W Y
C W C W C W
W Y W Y W Y
C W C W C W
W Y W Y W Y
C W C W C W
W Y W Y W Y
C W C W C W

The filters are white (no filter), yellow (lets in green and red),
and cyan (lets in green and blue). This is "demosaiced" by the
formula:

Red = W-C
Green = C+Y-W
Blue = W-Y

Where W, Y, and C are the white, yellow, and cyan sensor values.
For each pixel, one of these will be a straight CCD value while the
others will be averages of the 4 neighbors.

Luminance is R+G+B = (W-C)+(C+Y-W)+(W-Y) = W. This essentially
means that maximum resolution is given to luminance, while chroma
has half its resolution. This is what jpeg compression does first
anyway, and for the good reason that human vision is more sensitive
to luminance resolution than chroma resolution.

The resolution qualities are essentially similar to Bayer pattern
RGB filtering.

The advantage, I think, is that it will capture more light than a
normal RGB filter. This should translate to better sensitivity and
lower noise, right?

Am I missing something? Is there some reason why traditional RGB
filters are superior to this?
--
Isaac Kuo

Click to expand...

If you look at the r, g, and b channel error ranges independently, then the advantage is not so obvious. However, if you look at the error as a 3 dimensional "shape", then there's a big difference between a flat "saucer" shape and a thin "spindle" shape and a big "ball" shape.

Despite being based on R, G, and B sensors in the human eye, human vision perception is more sensitive to R+G+B (luminance) than any of the channels individually. I don't really understand it myself, but I understand the gyst of it means that the 3D "shape" of the ideal error range around a particular RGB color is a flattenned "blob" squeezed roughly along the R+G+B direction.

With equal amounts of traditional R, G, and B sensors, the "shape" of the error is a sphere. With the Bayer pattern of RGGB, this sphere is flattened into a blob--squeezed along the G direction. This is an improvement, but my intuitive feel is that my proposed approach is even better.

I'm not sure I really understand the issues correctly, though.

Anyway, there is some sort of "green bias" present in my proposed filter. With what I'm describing, green light is absorbed by all 4 out of 4 CCD elements. Red and blue light are each absorbed by only 3 out of 4 CCD elements. I don't really know exactly what this means, but it's got to somehow favor green.
--
Isaac Kuo

Very interesting discussion. Since 4 Megapixel sensors are so cheap now, why not use 3 in one camera? Bend the incoming light with a prism (ala Oly SLR) and allow each sensor to collect one color of data only. Combine the 3 and have a system similar to Foveon.

Ed Gelasi (Definitely NOT a rocket scientist)

bruce komusin said:

see
http://www.jvc.com/product.jsp?modelId=MODL026045&page=2
for example on model GR-DV2000U.

-bruce

IsaacKuo said:

Another Sensor idea?

It occurs to me that the current sensors filter away more light
than they really have to. Instead of the Bayer pattern RGB filter,
how about this alternate filter:

W Y W Y W Y
C W C W C W
W Y W Y W Y
C W C W C W
W Y W Y W Y
C W C W C W
W Y W Y W Y
C W C W C W

The filters are white (no filter), yellow (lets in green and red),
and cyan (lets in green and blue). This is "demosaiced" by the
formula:

Red = W-C
Green = C+Y-W
Blue = W-Y

Where W, Y, and C are the white, yellow, and cyan sensor values.
For each pixel, one of these will be a straight CCD value while the
others will be averages of the 4 neighbors.

Luminance is R+G+B = (W-C)+(C+Y-W)+(W-Y) = W. This essentially
means that maximum resolution is given to luminance, while chroma
has half its resolution. This is what jpeg compression does first
anyway, and for the good reason that human vision is more sensitive
to luminance resolution than chroma resolution.

The resolution qualities are essentially similar to Bayer pattern
RGB filtering.

The advantage, I think, is that it will capture more light than a
normal RGB filter. This should translate to better sensitivity and
lower noise, right?

Am I missing something? Is there some reason why traditional RGB
filters are superior to this?
--
Isaac Kuo

Click to expand...

--

Click to expand...

Hi Ed:

I think that is exactly what the original Foveon camera did. The alignment problems were a nightmare, if I recall, and the thing was large and tethered only.............

kunza

Ed G said:

Very interesting discussion. Since 4 Megapixel sensors are so cheap
now, why not use 3 in one camera? Bend the incoming light with a
prism (ala Oly SLR) and allow each sensor to collect one color of
data only. Combine the 3 and have a system similar to Foveon.

Click to expand...

Dave:

Like every bad idea, it sounded too easy! Although, technology may have shrunk things down a bit since then.

Ed

Dave Kunze said:

kunza

Ed G said:

Very interesting discussion. Since 4 Megapixel sensors are so cheap
now, why not use 3 in one camera? Bend the incoming light with a
prism (ala Oly SLR) and allow each sensor to collect one color of
data only. Combine the 3 and have a system similar to Foveon.

Click to expand...

Click to expand...

IsaacKuo said:

Another Sensor idea?

It occurs to me that the current sensors filter away more light
than they really have to. Instead of the Bayer pattern RGB filter,
how about this alternate filter:

W Y W Y W Y
C W C W C W
W Y W Y W Y
C W C W C W
W Y W Y W Y
C W C W C W
W Y W Y W Y
C W C W C W

Click to expand...

Why WASTE our time even discussing this.

I prefer to concentrate on the X3 type of technology. Any type of moasic pattern cannot help but be a magnet for moire artifacts. And unilizing 3 sensors with mirror/prisms is expensive and inherently not as "rugged" as a simpler optics system.

So I am hoping - PRAYING that Foveon, (or anyone else), DOES SUCCEED with that technology.

--
Thanks for reading .... JoePhoto

( Do You Ever STOP to THINK --- and FORGET to START Again ??? )

The big problem with adding 'luminance' cells is that your 'W' cell is going to be capturing two times more light than the coloured pixels. While that is great from a noise perspective, it causes one of two problems. For a normal sensor, your white pixels will be saturated much quicker (saturation is the cause of the native ISO of the sensor). The only way to avoid that saturation is that you have to reduce the fill rate (ie the percentage of the cell that is actually used for light capture) of each pixel - meaning either lower resolution (larger pixels with the same size active area) or more noise (same size pixels with a smaller active area).

Another problem, when using CCD sensors, is that you would really slow down the readoff process if you had to toggle the amplification for every single pixel (since you are alternating between white and colour sensors as they are read out). RGGB and CYYM sensors can opperate without changing the amplification for the entire readout process.

yeah people! stop thinking for yourselves and just follow what everone else is doing. now joe doesnt that sound kind of sad? just because your not innovative and a free thinker does not mean other people need to stop thinking as well. i guess its like your signature says.

now personally im looking foward to some analog cameras!! bring back the tube!!!

eric

JoePhoto said:

IsaacKuo said:

Another Sensor idea?

It occurs to me that the current sensors filter away more light
than they really have to. Instead of the Bayer pattern RGB filter,
how about this alternate filter:

W Y W Y W Y
C W C W C W
W Y W Y W Y
C W C W C W
W Y W Y W Y
C W C W C W
W Y W Y W Y
C W C W C W

Click to expand...

Why WASTE our time even discussing this.

I prefer to concentrate on the X3 type of technology. Any type of
moasic pattern cannot help but be a magnet for moire artifacts.
And unilizing 3 sensors with mirror/prisms is expensive and
inherently not as "rugged" as a simpler optics system.

So I am hoping - PRAYING that Foveon, (or anyone else), DOES
SUCCEED with that technology.

--
Thanks for reading .... JoePhoto

( Do You Ever STOP to THINK --- and FORGET to START Again ??? )

Click to expand...

yeah people! stop thinking for yourselves and just follow what everone else is doing. now joe doesnt that sound kind of sad? just because your not innovative and a free thinker does not mean other people need to stop thinking as well. i guess its like your signature says.

now personally im looking foward to some analog cameras!! bring back the tube!!!

eric

JoePhoto said:

IsaacKuo said:

Another Sensor idea?

It occurs to me that the current sensors filter away more light
than they really have to. Instead of the Bayer pattern RGB filter,
how about this alternate filter:

W Y W Y W Y
C W C W C W
W Y W Y W Y
C W C W C W
W Y W Y W Y
C W C W C W
W Y W Y W Y
C W C W C W

Click to expand...

Why WASTE our time even discussing this.

I prefer to concentrate on the X3 type of technology. Any type of
moasic pattern cannot help but be a magnet for moire artifacts.
And unilizing 3 sensors with mirror/prisms is expensive and
inherently not as "rugged" as a simpler optics system.

So I am hoping - PRAYING that Foveon, (or anyone else), DOES
SUCCEED with that technology.

--
Thanks for reading .... JoePhoto

( Do You Ever STOP to THINK --- and FORGET to START Again ??? )

Click to expand...

I think you may have misunderstood what I meant. I am indeed thinking "forward" to where we do not have a "mosiac-pattern" of any kind. The X3 idea is to eliminate the pattern which produces moire artifacts.

Note that I am not limiting that to the "Foveon-X3" ... cause maybe they are not doing it the "best" way. But I am saying that direction at least is where we need to be heading ... and not "backward".

eric said:

now personally im looking foward to some analog cameras!! bring
back the tube!!!

eric

JoePhoto said:

IsaacKuo said:

Another Sensor idea?

It occurs to me that the current sensors filter away more light
than they really have to. Instead of the Bayer pattern RGB filter,
how about this alternate filter:

W Y W Y W Y
C W C W C W
W Y W Y W Y
C W C W C W
W Y W Y W Y
C W C W C W
W Y W Y W Y
C W C W C W

Click to expand...

Why WASTE our time even discussing this.

I prefer to concentrate on the X3 type of technology. Any type of
moasic pattern cannot help but be a magnet for moire artifacts.
And unilizing 3 sensors with mirror/prisms is expensive and
inherently not as "rugged" as a simpler optics system.

So I am hoping - PRAYING that Foveon, (or anyone else), DOES
SUCCEED with that technology.

--
Thanks for reading .... JoePhoto

( Do You Ever STOP to THINK --- and FORGET to START Again ??? )

Click to expand...

Click to expand...

--
Thanks for reading .... JoePhoto

( Do You Ever STOP to THINK --- and FORGET to START Again ??? )

The Foveon or any other sensor that uses REGULAR PATTERN of sensors will always exhibit moiré pattern. Just because three colors are collected at one site does not eliminate that. It is the interfertence between regular pattern on image vs. regular pattern of sensors that causes moiré. Now if someone could invent a random-placement or irregular-pattern sensor array like human eye...

-bruce

JoePhoto said:

Note that I am not limiting that to the "Foveon-X3" ... cause maybe
they are not doing it the "best" way. But I am saying that
direction at least is where we need to be heading ... and not
"backward".

eric said:

now personally im looking foward to some analog cameras!! bring
back the tube!!!

eric

JoePhoto said:

IsaacKuo said:

Another Sensor idea?

It occurs to me that the current sensors filter away more light
than they really have to. Instead of the Bayer pattern RGB filter,
how about this alternate filter:

W Y W Y W Y
C W C W C W
W Y W Y W Y
C W C W C W
W Y W Y W Y
C W C W C W
W Y W Y W Y
C W C W C W

Click to expand...

Why WASTE our time even discussing this.

I prefer to concentrate on the X3 type of technology. Any type of
moasic pattern cannot help but be a magnet for moire artifacts.
And unilizing 3 sensors with mirror/prisms is expensive and
inherently not as "rugged" as a simpler optics system.

So I am hoping - PRAYING that Foveon, (or anyone else), DOES
SUCCEED with that technology.

--
Thanks for reading .... JoePhoto

( Do You Ever STOP to THINK --- and FORGET to START Again ??? )

Click to expand...

Click to expand...

--
Thanks for reading .... JoePhoto

( Do You Ever STOP to THINK --- and FORGET to START Again ??? )

Click to expand...

Ok so your W cell would be a ND filter...

Thomas Sapiano said:

The big problem with adding 'luminance' cells is that your 'W' cell
is going to be capturing two times more light than the coloured
pixels. While that is great from a noise perspective, it causes
one of two problems. For a normal sensor, your white pixels will
be saturated much quicker (saturation is the cause of the native
ISO of the sensor). The only way to avoid that saturation is that
you have to reduce the fill rate (ie the percentage of the cell
that is actually used for light capture) of each pixel - meaning
either lower resolution (larger pixels with the same size active
area) or more noise (same size pixels with a smaller active area).

Another problem, when using CCD sensors, is that you would really
slow down the readoff process if you had to toggle the
amplification for every single pixel (since you are alternating
between white and colour sensors as they are read out). RGGB and
CYYM sensors can opperate without changing the amplification for
the entire readout process.

Click to expand...

IsaacKuo said:

....
Luminance is R+G+B = (W-C)+(C+Y-W)+(W-Y) = W.

Click to expand...

Subtraction does nasty things with noise. If each of those sensors has an error(noise) term of n, I think the noise in the Luminance would be sqrt(7) n. If they are all added (absurd results, but ignore that for the moment), the signal goes up by a factor of seven, so the signal/noise ratio is decreased.

When there is subtraction, the noise increases, but the signal does not increase. Likely the increase would not be as simple as a factor of sqrt(7), but there would for sure be an increase.

bruce komusin said:

The Foveon or any other sensor that uses REGULAR PATTERN of sensors
will always exhibit moiré pattern. Just because three colors are
collected at one site does not eliminate that. It is the
interfertence between regular pattern on image vs. regular pattern
of sensors that causes moiré. Now if someone could invent a
random-placement or irregular-pattern sensor array like human eye...

-bruce

Click to expand...

OK ... you are correct in that any "regular" pattern of dots or lines can produce a "morie" artifact.

BUT; at least with the X3 type of sensor ... it won't be a MULTI-COLOR-RAINBOW type artifact.

The same with the "jiggles" on the edge of diagonal lines. We can only reduce/eliminate those with a "lot" of pixels ... but again ... with todays sensors those jiggles are MULTi-COLORED. That will not be the case with X3 type technology.

bruce komusin said:

JoePhoto said:

Note that I am not limiting that to the "Foveon-X3" ... cause maybe
they are not doing it the "best" way. But I am saying that
direction at least is where we need to be heading ... and not
"backward".

eric said:

now personally im looking foward to some analog cameras!! bring
back the tube!!!

eric

JoePhoto said:

IsaacKuo said:

Another Sensor idea?

It occurs to me that the current sensors filter away more light
than they really have to. Instead of the Bayer pattern RGB filter,
how about this alternate filter:

W Y W Y W Y
C W C W C W
W Y W Y W Y
C W C W C W
W Y W Y W Y
C W C W C W
W Y W Y W Y
C W C W C W

Click to expand...

Why WASTE our time even discussing this.

I prefer to concentrate on the X3 type of technology. Any type of
moasic pattern cannot help but be a magnet for moire artifacts.
And unilizing 3 sensors with mirror/prisms is expensive and
inherently not as "rugged" as a simpler optics system.

So I am hoping - PRAYING that Foveon, (or anyone else), DOES
SUCCEED with that technology.

--
Thanks for reading .... JoePhoto

( Do You Ever STOP to THINK --- and FORGET to START Again ??? )

Click to expand...

Click to expand...

--
Thanks for reading .... JoePhoto

( Do You Ever STOP to THINK --- and FORGET to START Again ??? )

Click to expand...

--

Click to expand...

--
Thanks for reading .... JoePhoto

( Do You Ever STOP to THINK --- and FORGET to START Again ??? )

Aaron van de Sande said:

Ok so your W cell would be a ND filter...

Click to expand...

Yes, that would solve the problem - however it removes the primary point of this configuration (to reduce noise by letting more light through to the sensor).