Final output of the Foveon X3 ?

[Ron Parr]

If I have my theoritical "white" wall with "red" dots/squares. I
am only interested in how many "red" sensors there are.

You can construct pathological images like this, but these will not reflect actual performance on real images.

Now you have the problem of interpolation. Yes, interpolation is
guessing information and it's impossible to be right all of the
time. However, it's also impossible to be wrong all of the time.
The actual performance is somewhere in the middle.

True .. but the same is true if you interpolate a X3 file to a
higher density. The fact remains that there are only 1.5m "groups"
of pixels able to produce "white" dots in a conventional 6mps
sensor ... there are 3.54m of those same "groups" in X3.

THE ONLY DIFFERENCE IS THEY ARE STACKED INSTEAD OF SIDE-by-SIDE.

Yes and no. You're sampling the luminance signal at a higher rate than the other channels with Bayer, but you're more sensitive to the luminance channel. Most of the time it will appear as good as if you sampled the luminance from every other pixel.

I'm not trying to argue against the X3 approach. It's just that you're oversimplifying things. The Bayer pattern was designed to do a good job of hiding the limitations of interpolation on the majority of images. There's no free lunch, but on many images it will do a better job of fooling you than the pixel count would suggest.

If you're expecting a 3X improvement with the Foveon chip, you'll be disappointed. I don't like the whole numbers game, but there's one thing I'm 100% confident of: a 3X improvement on typical images is too much to hope for.

--
Ron Parr
FAQ: http://www.cs.duke.edu/~parr/photography/faq.html
Gallery: http://www.pbase.com/parr/

 

[JoePhoto]

If I have my theoritical "white" wall with "red" dots/squares. I
am only interested in how many "red" sensors there are.

You can construct pathological images like this, but these will not
reflect actual performance on real images.

Now you have the problem of interpolation. Yes, interpolation is
guessing information and it's impossible to be right all of the
time. However, it's also impossible to be wrong all of the time.
The actual performance is somewhere in the middle.

True .. but the same is true if you interpolate a X3 file to a
higher density. The fact remains that there are only 1.5m "groups"
of pixels able to produce "white" dots in a conventional 6mps
sensor ... there are 3.54m of those same "groups" in X3.

THE ONLY DIFFERENCE IS THEY ARE STACKED INSTEAD OF SIDE-by-SIDE.

Yes and no. You're sampling the luminance signal at a higher rate
than the other channels with Bayer, but you're more sensitive to
the luminance channel. Most of the time it will appear as good as
if you sampled the luminance from every other pixel.

I'm not trying to argue against the X3 approach. It's just that
you're oversimplifying things. The Bayer pattern was designed to
do a good job of hiding the limitations of interpolation on the
majority of images. There's no free lunch, but on many images it
will do a better job of fooling you than the pixel count would
suggest.

If you're expecting a 3X improvement with the Foveon chip, you'll
be disappointed. I don't like the whole numbers game, but there's
one thing I'm 100% confident of: a 3X improvement on typical
images is too much to hope for.

But a 3X improvement over "what" is too much to hope for ???

I indeed DO expect a 3X improvement over a "conventional" 3mps sensor. (with NO color-artifacting from interpolation).

And I expect a 2X improvement over conventional 6mps sensors. (since they only have 1.5mps "groups" compared to the 3mps "groups" of X3)

And I personally would not be surprised if it closely equals Canons new 11mps ... and Kodaks 14mps.

Will I be right ??? .... Heck I don't know. I realize there are many other factors that affect image quality, (resolution/noise/density-range/iso speed).

But you know ... there is even a possibility that the overall quality will be even better than the 11-14mps if there is no "color-morie".

I am also concerned about Shutter Lag ... and the relatively slow ISO 400 speed limitation; (I would prefer a top ISO of 100,000)

--
Ron Parr
FAQ: http://www.cs.duke.edu/~parr/photography/faq.html
Gallery: http://www.pbase.com/parr/

--
Thanks for reading .... JoePhoto

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

 

[Ron Parr]

But a 3X improvement over "what" is too much to hope for ???

I indeed DO expect a 3X improvement over a "conventional" 3mps
sensor. (with NO color-artifacting from interpolation).

3X improvement over the equivalent size Bayer pattern sensor is too much to hope for on most images. The green channel is grabbing a good chunk of the luminance signal, which is the predominant influence on your perception of detail in most images. A Bayer pattern sensor is half green.

--
Ron Parr
FAQ: http://www.cs.duke.edu/~parr/photography/faq.html
Gallery: http://www.pbase.com/parr/

 

[David Barkin]

Hi Joe

Foveon (X3) faced a decision. They could very easily have simpy
called their chip a 10.62 mps and we would not be "arguing" about
it.

(They could even have gone a step further and say it "equals" a
Bayer 14.16mps chip since they have R-G-B-B)

But instead they made a tactical decison to help illustrate their
"different" technology by adopting an entirely new rating system
that is indeed MORE ACCURATE and HONEST.

To illustrate my point ... imagine for a minute a white wall with
RED DOTS. Not imagine the density of those dots exactely equals
3.54mps.

If you photographic that wall with a 6mps sensor ... there are TWO
VERY DIFFERENT POSSIBILITIES ... one is that the red "dots"
exactely align with the RED-SENSORS ....

another possibility is that they do not align with any of the red
sensors.

However ... EVEN IF they DO "align" perfectly ... you will NOT GET
A PERFECT "RED" DOT .. because the "interpolation" will mix in the
"white" from the adjoining sensors and reduce its brilliance to a
lighter shade of "pink".

The second possibility, (and 3X as likely since there is only 1/4
"red" sensors); of NONE of the red dots aliging with the red
sensors produces an almost HORRIFYING RESULT. I could not even
attempt to tell you what would happen.

HOWEVER ... that same scenario on a X3 sensor would produce a
PERFECT IMAGE of "RED" dots.

This is an interesting idea. Have yu put it to the test? Why not? My dog is a poor subject. I live in an apartment. Can anyone set such a test up to see what would happen?

Can you postulate an exact test? (i.e. lens, dot size, distance, etc) or am I completely missing the point?

Dave

 

[JoePhoto]

Foveon (X3) faced a decision. They could very easily have simpy
called their chip a 10.62 mps and we would not be "arguing" about
it.

(They could even have gone a step further and say it "equals" a
Bayer 14.16mps chip since they have R-G-B-B)

But instead they made a tactical decison to help illustrate their
"different" technology by adopting an entirely new rating system
that is indeed MORE ACCURATE and HONEST.

To illustrate my point ... imagine for a minute a white wall with
RED DOTS. Not imagine the density of those dots exactely equals
3.54mps.

If you photographic that wall with a 6mps sensor ... there are TWO
VERY DIFFERENT POSSIBILITIES ... one is that the red "dots"
exactely align with the RED-SENSORS ....

another possibility is that they do not align with any of the red
sensors.

However ... EVEN IF they DO "align" perfectly ... you will NOT GET
A PERFECT "RED" DOT .. because the "interpolation" will mix in the
"white" from the adjoining sensors and reduce its brilliance to a
lighter shade of "pink".

The second possibility, (and 3X as likely since there is only 1/4
"red" sensors); of NONE of the red dots aliging with the red
sensors produces an almost HORRIFYING RESULT. I could not even
attempt to tell you what would happen.

HOWEVER ... that same scenario on a X3 sensor would produce a
PERFECT IMAGE of "RED" dots.

This is an interesting idea. Have yu put it to the test? Why not?
My dog is a poor subject. I live in an apartment. Can anyone set
such a test up to see what would happen?

Can you postulate an exact test? (i.e. lens, dot size, distance,
etc) or am I completely missing the point?

Dave

Well .... I suppose that could be calculated .... but the density of the dot-pattern and lens-distance would factor together to produce the exact 3.54mps of "red" dots.

But actually we CANNOT duplicate it anyway because it would require a 14.16mps sensor to also have 3.54mps of "red" only sensors.

And of course from a practical standpoint we could never take photos where "all" of the red pixels directly impinged on only the red sensors ... nor one where "none" of them did.

So I was only trying to use the illustration to make people "think" of what is happening with interpolation. We like to think of it involving a "mixture" of all colors ... but actually ... this is just as valid.

BUT... BUT ... David ... THANK YOU for posting this reply ... BUT, BUT ... this is NOT the post I meant. I "duplicated" this post in a NEW THREAD to get a discussion going ... since this is so far near the bottom of this one. So it was the "NEW" thread I wanted to see what could be gererated.

So can you PLEASE find the "new" thread ... it is probably now on at least the "second" page of the News-forum ... the title of the "new" thread is ... "Why X3 should be (is) so good" .. and make "any" reply there ... hopefully that will put it back on "top" of the scrowl and lets see if it generates some interest/discussion.

Thank You

--
Thanks for reading .... JoePhoto

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

 

[rick decker]

David: I am not clear...did you print it at 16x20. I used SI to re-size and I agree with you on how great it looks and how sharp. There haved been a number of postings (including mine) on 12x18 and how great they look but since poeple can't see them, they are not necessarily believers. I wish I could do 16x20.

Hi All

I just downloaded one of the new samples that Phill posted. I
downloaded the full sized version, although of course it was in
JPEG and only 1.4 MB.

At any rate the image compared quite favoribly with the best of my
shots with the D1x. Of course no one posts bad samples, so in and
of itself this really doesn't mean to much. I then interpolated the
image up, doubled it, and here the results were amazing. It was if
I was looking at an original image not an interpolation.

Now I routinly interpolate up for printing at 16 x 20 and the
results are just fine, but this was crystal clear, truly amazing.

Dave

 

[David Barkin]

Can't find any thread you started with that title, neither in news or open talk.

Start it again in Open talk - It's not a Sigma issue, rather a sensor issue.

Dave

BUT... BUT ... David ... THANK YOU for posting this reply ... BUT,
BUT ... this is NOT the post I meant. I "duplicated" this post in
a NEW THREAD to get a discussion going ... since this is so far
near the bottom of this one. So it was the "NEW" thread I wanted
to see what could be gererated.

So can you PLEASE find the "new" thread ... it is probably now on
at least the "second" page of the News-forum ... the title of the
"new" thread is ... "Why X3 should be (is) so good" .. and make
"any" reply there ... hopefully that will put it back on "top" of
the scrowl and lets see if it generates some interest/discussion.

Thank You

--
Thanks for reading .... JoePhoto

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

 

[Matthew Cromer]

It is impossible for a Bayer camera to measure luminance precisely at each pixel location.

You cannot accurately measure luminance at each pixel location when you are filtering out different colors at different positions. Each filter removes a large portion of the light of different frequencies. Hence the photosites cannot accurately measure overall luminance.

--
my favorite work: http://www.pbase.com/sdaconsulting/favorite_work

 

[David Barkin]

Hi Rick

David: I am not clear...did you print it at 16x20. I used SI to
re-size and I agree with you on how great it looks and how sharp.
There haved been a number of postings (including mine) on 12x18 and
how great they look but since poeple can't see them, they are not
necessarily believers. I wish I could do 16x20.

The damn stuff is just to expensive for me to print out a 16 x 20 just for the heck of it. What I did do was set up a 16 x 20 image and print a crop from that.

Yes the results are better.

However today's home printers really can't take full advantage of this. If you're getting marvelous home prints now, what's the big deal?

I see two advantages.

1. After you take an image and you CROP it very small, it can be interpolated up and THEN produce noticably better results.

2. For professional use on high dpi printing runs, the difference should be quite clear.

Dave

 

[David Barkin]

Hi Rick

I'm still playing around with these samples. Now when I make a large print I interpolate up. I interpolate up if for no other reason to eliminate pixalation, jaggies. Now the difference between the Foveon samples and the D1x images is that interpolating the D1x does indeed smooth the image. Increasing the size of the image will also reveal details lost because of size. But these details on D1x shots are foggy, blurred.

This is the one of the big differences with these Foveon samples. The interpolated images show these details crytal clear - no fog (I'm sure these a technical term for this, but I don't know it).

Once again this is much more like working with high resoltion scans from film then from digital.

So what I've been doing now is making .3 inch crops and interpolating them up to 3 x 5, once again I find the results astounding. The interpolation is by a factor of 10. So my final prints are 10 Megs.

Dave

David: I am not clear...did you print it at 16x20. I used SI to
re-size and I agree with you on how great it looks and how sharp.
There haved been a number of postings (including mine) on 12x18 and
how great they look but since poeple can't see them, they are not
necessarily believers. I wish I could do 16x20.

The damn stuff is just to expensive for me to print out a 16 x 20
just for the heck of it. What I did do was set up a 16 x 20 image
and print a crop from that.

Yes the results are better.

However today's home printers really can't take full advantage of
this. If you're getting marvelous home prints now, what's the big
deal?

I see two advantages.

1. After you take an image and you CROP it very small, it can be
interpolated up and THEN produce noticably better results.

2. For professional use on high dpi printing runs, the difference
should be quite clear.

Dave

 

[rick decker]

I tried one crop on the sushi and interpolated up and did an 8x12 but in calculating I think it was equivalent only to a 15x22 with the whole image. I will try something a little smaller. I have been printing at about 270 dpi on my 1270 which I think should be adequate.

rick

I'm still playing around with these samples. Now when I make a
large print I interpolate up. I interpolate up if for no other
reason to eliminate pixalation, jaggies. Now the difference between
the Foveon samples and the D1x images is that interpolating the D1x
does indeed smooth the image. Increasing the size of the image will
also reveal details lost because of size. But these details on D1x
shots are foggy, blurred.

This is the one of the big differences with these Foveon samples.
The interpolated images show these details crytal clear - no fog
(I'm sure these a technical term for this, but I don't know it).

Once again this is much more like working with high resoltion scans
from film then from digital.

So what I've been doing now is making .3 inch crops and
interpolating them up to 3 x 5, once again I find the results
astounding. The interpolation is by a factor of 10. So my final
prints are 10 Megs.

Dave

David: I am not clear...did you print it at 16x20. I used SI to
re-size and I agree with you on how great it looks and how sharp.
There haved been a number of postings (including mine) on 12x18 and
how great they look but since poeple can't see them, they are not
necessarily believers. I wish I could do 16x20.

The damn stuff is just to expensive for me to print out a 16 x 20
just for the heck of it. What I did do was set up a 16 x 20 image
and print a crop from that.

Yes the results are better.

However today's home printers really can't take full advantage of
this. If you're getting marvelous home prints now, what's the big
deal?

I see two advantages.

1. After you take an image and you CROP it very small, it can be
interpolated up and THEN produce noticably better results.

2. For professional use on high dpi printing runs, the difference
should be quite clear.

Dave

 

[Patrick S Pilkenton]

I took the sample to 40 x 60 then out of that I cut out the eye at 4 x 6 and it was still clear. The 40 x 60 products a 115meg file