Caution about X3

Leon Wittwer · Feb 20, 2002

If you consider the technical definition of resolving power, that is, how close can two points be and still be resolved, the Fovean should be able to resolve two points at 0.707 times the distance of a comparable Bayer CCD with the same number of pixels and the same size CCD. This assumes that most of the relevant info is in the green channel. If I have a 3.5MP Fovean, it would take a 7MP (approximately) to get the same resolving power. It is too bad that folks don't use resolving power (lines per inch or equivalent) as a measure (like in Phil's reviews) rather than the total number of pixels (aka "resolution").

Leon Wittwer · Feb 20, 2002

Gary, If you buy my message above, the implication is that the 3.5 MP Fovean would print approximately as well as the 6 MP Bayer ccd. If you go below 200, then you upsample to keep the printer happy. Epson printers (aka my 1270 and 2000P) print at about 150 lines per inch or 300 pixels per inch max resolving power. I measured this directly with some tests. On the other hand, I see improvement in image quality (with a bit of magnification) as I print up to about 400+ pixels inch native out of my digital camera. How can this be? It works because the true resolving power (point to point) of my Bayer ccd is about 1.414 times worse than having a true measured green (mostly) at each pixel. Thus it takes about 1.4*300 pixels per inch in the printed image to max out the 300 pixels per inch in the printer. Practical realities decrease this a bit but you get the idea. The Fovean in principle should max out the 300 ppi printer with about 300 ppi.

Gary Eickmeier said:
That's fine, but when I go to print the image, I can only go so
large before I am printing at below 200ppi. It just seems that I
would be better off with a 6mp regular CCD than a 3.5mp X3. Of
course, the ideal would be a 6mp X3, but the story is that we can
multiply the resolution by 3 with the "X3" and get the true
resolution of the imager, which is what I am struggling with.

Gary Eickmeier

Hans Keesom · Feb 20, 2002

Yes the number of pixels are the same,BUT!!!!! the pixels are different

Until now each pixel would only catch one color and pixels where assigned to one color in a RGBG or likewise pattern. Therefore, in color-terms they were only at 1/3 or 1/4 of the advertised resolution. Using guessing and interpolation calculations the camera outputted a picture at the advertised resolution(3,5 mp).

X3 means each advertised pixel is really Red Green and Blue and no guessing is done. The quality of the outputted picture should therefore in general be much, much better.

Problably you can upscale a 3.5 mp X3 picture to 10.5 mp and have the same quality as a 10,5 mp non-x3 camera.

Steven Lyons said:
There seems to be alot of confusion about this. Maybe Phil could
weigh in. It appears that what you end up with is only a 3.5 MP
image, so the pixel dimensions (and resultant ability to enlarge or
reproduce in print) will be the same as with any other camera with
a chip this size. The images appear to be very clean and each pixel
appears to be very color accurate in the sample shots, so this may
give you more headroom for enlargement. A 6MP CCD might be able to
compete even if its noisier.

The other question I have not seen addressed is that of the speed
of capture. The Sigma camera apparently won't do continuous
shooting at full rez and there is no mention of the buffer size in
the specs(is there?). Wonder how long you have to wait between
shots.
--
Steven Lyons
http://stevenlyons.com

Gary Eickmeier · Feb 20, 2002

Leon Wittwer said:
Gary, If you buy my message above, the implication is that the 3.5
MP Fovean would print approximately as well as the 6 MP Bayer ccd.
If you go below 200, then you upsample to keep the printer happy.
Epson printers (aka my 1270 and 2000P) print at about 150 lines per
inch or 300 pixels per inch max resolving power. I measured this
directly with some tests. On the other hand, I see improvement in
image quality (with a bit of magnification) as I print up to about
400+ pixels inch native out of my digital camera. How can this be?
It works because the true resolving power (point to point) of my
Bayer ccd is about 1.414 times worse than having a true measured
green (mostly) at each pixel. Thus it takes about 1.4*300 pixels
per inch in the printed image to max out the 300 pixels per inch
in the printer. Practical realities decrease this a bit but you
get the idea. The Fovean in principle should max out the 300 ppi
printer with about 300 ppi.

So you can multiply the Foveon's resolution by some factor if you upsample its output? Where does that leave us? You can upsample anything. Doesn't make it so.

What this says to me is that you can take a 3.3mp Foveon image and upsample it to gain back the resolution to equal the Bayer CCD, but when you do that, you make it guess about the colors of adjacent pixels, so we're back where we started!

I will concede the absolute superiority of a Foveon 3.3mp imager to a Bayer pattern 3.3mp imager, but I question the "multiplication factor" capability of the Foveon when comparing resolution to a normal imager. A 3.3mp image is a 3.3mp image, even if the color resolution is greater.

Gary Eickmeier

Gary Eickmeier · Feb 20, 2002

Paul W. Walters said:
Gary,

Listen... The story is NOT that you can multiply it by three!
Come on, read back on this very thread. It's just 3.5 megapixels
of BETTER quality pixels. We've yet to see how well the improved
colors and lack of noise compares to the larger CCDs, but the story
IS NOT 3.5 x 3 = 9. It's 3.5 of GREAT QUALITY pixels!

Clearly, on Phil's description of the Sigma SD-9 and on Foveon's website, it has wording like

" Pre-PMA 2002: Sigma's new SD9 D-SLR will be the first digital camera to utilize Foveon's recently announced X3 sensor technology. This new camera has a 3.43 megapixel (effective) X3 sensor which outputs 2268 x 1512 x 3 pixels. Each pixel is 9 microns which makes the sensor 20.7 x 13.8 mm (a 1.7x focal length multiplier). The SD9 has a Sigma lens mount and a unique dust protector just behind the lens mount to stop dust from attaching itself to the sensor. Price is expected to be around US$3,000."

Now, is there something about this statement that I am misinterpreting?

Gary Eickmeier

Hans Keesom · Feb 20, 2002

Gary, maybe this will help

Yes the number of pixels are the same,BUT!!!!! the pixels are different

Until now each pixel would only catch one color and pixels where assigned to one color in a RGBG or likewise pattern. Therefore, in color-terms they were only at 1/3 or 1/4 of the advertised resolution. Using guessing and interpolation calculations the camera outputted a picture at the advertised resolution(3,5 mp).

X3 means each advertised pixel is really Red Green and Blue and no guessing is done. The quality of the outputted picture should therefore in general be much, much better.

Problably you can upscale a 3.5 mp X3 picture to 10.5 mp and have the same quality as a 10,5 mp non-x3 camera.

Gary Eickmeier said:
Paul W. Walters said:

Gary,

Listen... The story is NOT that you can multiply it by three!
Come on, read back on this very thread. It's just 3.5 megapixels
of BETTER quality pixels. We've yet to see how well the improved
colors and lack of noise compares to the larger CCDs, but the story
IS NOT 3.5 x 3 = 9. It's 3.5 of GREAT QUALITY pixels!

Click to expand...

Clearly, on Phil's description of the Sigma SD-9 and on Foveon's
website, it has wording like

" Pre-PMA 2002: Sigma's new SD9 D-SLR will be the first digital
camera to utilize Foveon's recently announced X3 sensor technology.
This new camera has a 3.43 megapixel (effective) X3 sensor which
outputs 2268 x 1512 x 3 pixels. Each pixel is 9 microns which makes
the sensor 20.7 x 13.8 mm (a 1.7x focal length multiplier). The SD9
has a Sigma lens mount and a unique dust protector just behind the
lens mount to stop dust from attaching itself to the sensor. Price
is expected to be around US$3,000."

Now, is there something about this statement that I am
misinterpreting?

Gary Eickmeier

Hans Keesom · Feb 20, 2002

Gary Eickmeier said:
A 3.3mp image is a 3.3mp image, even if the color
resolution is greater.

The number of pixels are the same,BUT!!!!! the pixels are different. What we used to call 3.3 mp isn't really 3.3 mp or only 3.3 m monocolorpixels! X3 has 3,3 m colorpixels!

Until now each pixel would only catch one color and pixels where assigned to one color in a RGBG or likewise pattern. Therefore, in color-terms they were only at 1/3 or 1/4 of the advertised resolution. Using guessing and interpolation calculations the camera outputted a picture at the advertised resolution(3,5 mp).

X3 means each advertised pixel is really Red Green and Blue and no guessing is done. The quality of the outputted picture should therefore in general be much, much better.

Problably you can upscale a 3.5 mp X3 picture to 10.5 mp and have the same quality as a 10,5 mp non-x3 camera.

Karl Guttag · Feb 20, 2002

Hans Keesom said:
Problably you can upscale a 3.5 mp X3 picture to 10.5 mp and have
the same quality as a 10,5 mp non-x3 camera.

This will depend on how you define "quality" and certainly will not be true in many cases. A 10.5MP Bayer pattern will be able to capture some fine detail that could not be captured with a 3.5MP X3. While the filters are colored, they are still getting some of the intensity information at full resolution.

Karl--Karl

Paul W. Walters · Feb 20, 2002

Gary,

That's all there is to it. It is a 3.5 megapixel sensor with far greater accuracy. It DOES enlarge better than a 3.5 megapixel CCD, but STILL when you start enlarging it you're going back to guesswork.

With my tests on the sample images, you can enlarge it through Genuine Fractals by up to 200% or so without it looking bad. But that's just in comparing it to a CCD. Once you enlarge it, it doesn't look as crisp as it looked at 100%. Eventually we won't use the CCD for comparison anymore and it'll all come back to the actual resolution. In this case, 3.5 megapixels. We just have 3.5 Accurate megapixels.

Paul

Gary Eickmeier said:
I will concede the absolute superiority of a Foveon 3.3mp imager to
a Bayer pattern 3.3mp imager, but I question the "multiplication
factor" capability of the Foveon when comparing resolution to a
normal imager. A 3.3mp image is a 3.3mp image, even if the color
resolution is greater.

Gary Eickmeier

Alistair60736 · Feb 20, 2002

But the bayer interpolation introduces aliasing noise into the picture when the algorithm makes a bad guess at value in between. This noise would actually obscure the detail that is there. So perhaps the answer is not that the resolution is better as such just that the resolution that is there is not as obscured by noise.

David SL said:
Gary,

The difference between the two techniques lies in the variance in
color in adjacent pixels. Humans percieve detail by noting
transitions in color, the more accurate the color the highier in
percieved resolution an image will appear to a human. The cones in
the eye that detect color are mostly responsible for our ability to
discern fine detail precisely because of their color sensitivity.

Gary Eickmeier · Feb 20, 2002

Hans Keesom said:
Gary, maybe this will help

Yes the number of pixels are the same,BUT!!!!! the pixels are
different

Until now each pixel would only catch one color and pixels where
assigned to one color in a RGBG or likewise pattern. Therefore, in
color-terms they were only at 1/3 or 1/4 of the advertised
resolution. Using guessing and interpolation calculations the
camera outputted a picture at the advertised resolution(3,5 mp).

X3 means each advertised pixel is really Red Green and Blue and no
guessing is done. The quality of the outputted picture should
therefore in general be much, much better.

Problably you can upscale a 3.5 mp X3 picture to 10.5 mp and have
the same quality as a 10,5 mp non-x3 camera.

OK, I guess you're going to have to define "quality" for me. If your problem with the Bayer is that adjacent pixels are guessing about color, what do you have to say about adjacent pixels in an upsampled image?

And who said that you can upsample a Foveon image to make it comparable to the Bayer? Their story is simply that the pixel count can be multiplied by 3 to give the resolution of an X3 imager. I don't understand that. There will still be 3.5 megapixels. No one said to upsample it.

If I am belaboring a point that no one had an answer to, then let's just wait until we can ask someone at PMA.

Thanks,

Gary Eickmeier

Bob the Printer · Feb 20, 2002

Gary Eickmeier said:
I just want to ask the assembled experts about the Foveon's
supposed "times 3" resolution factor when specifying the pixel
count. For example, they say their 3.3mp Sigma will have resolution

What they're really referring to is 'color resolution'. Take this extreme example for instance..

Suppose you have a subject which fills three pixels across (at the image plane) and that subject is red/blue/green (in other words, one pixel is red, one is blue, and the third is green)..

Now, a bayer pattern sensor would not be able to resolve these three colors, but the X3 sensor most likely would be able to resolve them..

Does that make sense?? It does to me but.....

--Sincerely,Bob the Printer

David SL · Feb 20, 2002

Alistair said:
But the bayer interpolation introduces aliasing noise into the
picture when the algorithm makes a bad guess at value in between.
This noise would actually obscure the detail that is there. So
perhaps the answer is not that the resolution is better as such
just that the resolution that is there is not as obscured by noise.

That's almost exactly what I stated in my post:

Alistair60736 said:
David SL wrote:

Color information "smears" across the image, this leads to a loss of human > percieved detail in comparison to X3 where the color "smear" does not > occur, detail is maintained, color accuracy is maintained and percieved > resolution goes(or rather stays..) up.

The color "smear"(what you called "noise" but technically its color variation due to inaccurate guesses at the color values at each pixel) induced by bayer interpolation reduces the percieved resolution of the image (relative to X3) even if the number of pixels between images are the same. Again, this is because humans use color variation to percieve detail, the more accurate color is the highier percieved detail.

Regards,

--DSL

Paul W. Walters · Feb 20, 2002

Who are They that say this? The x3 doesn't mean "TIMES 3" it means "x3 technology" -- x3 technology is just a SINGLE PIXEL being able to sense THREE DIFFERENT COLORS. I've never seen Fovean say that you need to multiply the resolution "times 3". They say that you have much more ability to enlarge an x3 image than a CCD image (and this is true), but they don't specify to multiply the resolution by three.

Paul

Gary Eickmeier said:
............................Their story is simply that the pixel count
can be multiplied by 3 to give the resolution of an X3 imager.> Gary Eickmeier

Bob the Printer · Feb 20, 2002

Gary Eickmeier said:
That's fine, but when I go to print the image, I can only go so
large before I am printing at below 200ppi. It just seems that I
would be better off with a 6mp regular CCD than a 3.5mp X3. Of
course, the ideal would be a 6mp X3, but the story is that we can
multiply the resolution by 3 with the "X3" and get the true
resolution of the imager, which is what I am struggling with.

Once again, it's not the total resolution that gets multiplied, it's the COLOR RESOLUTION..... See my previous example to see why..

--Sincerely,Bob the Printer

Gary Eickmeier · Feb 20, 2002

Paul W. Walters said:
Who are They that say this? The x3 doesn't mean "TIMES 3" it means
"x3 technology" -- x3 technology is just a SINGLE PIXEL being able
to sense THREE DIFFERENT COLORS. I've never seen Fovean say that
you need to multiply the resolution "times 3". They say that you
have much more ability to enlarge an x3 image than a CCD image (and
this is true), but they don't specify to multiply the resolution by
three.

I'll just repeat my post I sent you about two posts up:

Clearly, on Phil's description of the Sigma SD-9 and on Foveon's website, it has wording like

" Pre-PMA 2002: Sigma's new SD9 D-SLR will be the first digital camera to utilize Foveon's recently announced X3 sensor technology. This new camera has a 3.3 megapixel (effective) X3 sensor which outputs 2268 x 1512 x 3 pixels. Each pixel is 9 microns which makes the sensor 20.7 x 13.8 mm (a 1.7x focal length multiplier). The SD9 has a Sigma lens mount and a unique dust protector just behind the lens mount to stop dust from attaching itself to the sensor. Price is expected to be around US$3,000."

Now, is there something about this statement that I am misinterpreting?

Gary Eickmeier

Bob the Printer · Feb 20, 2002

Gary Eickmeier said:
Clearly, on Phil's description of the Sigma SD-9 and on Foveon's
website, it has wording like

" Pre-PMA 2002: Sigma's new SD9 D-SLR will be the first digital
camera to utilize Foveon's recently announced X3 sensor technology.
This new camera has a 3.43 megapixel (effective) X3 sensor which
outputs 2268 x 1512 x 3 pixels. Each pixel is 9 microns which makes
the sensor 20.7 x 13.8 mm (a 1.7x focal length multiplier). The SD9
has a Sigma lens mount and a unique dust protector just behind the
lens mount to stop dust from attaching itself to the sensor. Price
is expected to be around US$3,000."

Now, is there something about this statement that I am
misinterpreting?

So where does it say that you multiply the total sensor resolution by 3?? It doesn't.
--Sincerely,Bob the Printer

Gary Eickmeier · Feb 20, 2002

Bob the Printer said:
Gary Eickmeier said:

I just want to ask the assembled experts about the Foveon's
supposed "times 3" resolution factor when specifying the pixel
count. For example, they say their 3.3mp Sigma will have resolution

Click to expand...

What they're really referring to is 'color resolution'. Take this
extreme example for instance..

Suppose you have a subject which fills three pixels across (at the
image plane) and that subject is red/blue/green (in other words,
one pixel is red, one is blue, and the third is green)..

Now, a bayer pattern sensor would not be able to resolve these
three colors, but the X3 sensor most likely would be able to
resolve them..

Does that make sense?? It does to me but.....

Yes, it makes terrific sense. The X3 has greater color resolution.

My question has to do with multiplying the number of pixels by 3 to get the "real" or "equivalent" pixels for the Foveon. You can multiply all you want, but it is still a 3.3mp sensor.

My caution is that people will think they are getting a 10mp sensor and they are not. The math is still the same for enlarging it to see how many ppi you would be printing at, and at what point pixelization would occur.

Is that some sort of esoteric point or something?

Gary Eickmeier

Gary Eickmeier · Feb 20, 2002

Bob the Printer said:
Gary Eickmeier said:

Clearly, on Phil's description of the Sigma SD-9 and on Foveon's
website, it has wording like

" Pre-PMA 2002: Sigma's new SD9 D-SLR will be the first digital
camera to utilize Foveon's recently announced X3 sensor technology.
This new camera has a 3.43 megapixel (effective) X3 sensor which
outputs 2268 x 1512 x 3 pixels. Each pixel is 9 microns which makes
the sensor 20.7 x 13.8 mm (a 1.7x focal length multiplier). The SD9
has a Sigma lens mount and a unique dust protector just behind the
lens mount to stop dust from attaching itself to the sensor. Price
is expected to be around US$3,000."

Now, is there something about this statement that I am
misinterpreting?

Click to expand...

So where does it say that you multiply the total sensor resolution
by 3?? It doesn't.

Now this is getting silly. Just read the thing. You want me to read it for you? It says

"This new camera has a 3.43 megapixel (effective) X3 sensor which outputs 2268 x 1512 x 3 pixels."

Do you see the "x 3" in the description of the number of pixels? It comes through clearly on my server.

Gary Eickmeier

Bob the Printer · Feb 20, 2002

Gary Eickmeier said:
My question has to do with multiplying the number of pixels by 3 to
get the "real" or "equivalent" pixels for the Foveon. You can
multiply all you want, but it is still a 3.3mp sensor.

My caution is that people will think they are getting a 10mp sensor
and they are not. The math is still the same for enlarging it to
see how many ppi you would be printing at, and at what point
pixelization would occur.

Is that some sort of esoteric point or something?

But where does it say that you muliply the sensor resolution by 3?? Nowhere that I've seen in the 'official literature'...

But consider the fact that a 3.5MP CCD sensor actually only has 1/4 of that resolution for each color (well 1/2 for green and 1/4 for red and blue).

The bottom line is that the X3 sensor offers much better pictures!

--Sincerely,Bob the Printer

Caution about X3

Forum Pro

Forum Pro

Hans Keesom

Guest

Senior Member

Senior Member

Hans Keesom

Guest

Hans Keesom

Guest

Senior Member

Senior Member

Forum Enthusiast

Senior Member

Leading Member

Senior Member

Senior Member

Leading Member

Senior Member

Leading Member

Senior Member

Senior Member

Leading Member

Similar threads

Keyboard shortcuts