The death of the Bayer filter? Maybe not.

Alfred Molon wrote:
Here is a shot I took today:
http://www.molon.de/Bayer_test.jpg

It's the Brandhorst museum in Munich, near the Pinakothek der Moderne:
http://www.museum-brandhorst.de

It's a perfect image to check the limits of the performance of a Bayer
sensor. Have a look at the wall facing left.

Firefox downsampling creates aliasing there (not visible at 100%).

Now we need a Foveon sensor with the same pixel count at the
same place and with the same lens to compare. Better hurry,
before the building completely falls over backwards.

-Wolfgang

On Sat, 05 May 2012 16:04:30 -0700, nospam
wrote:

In article , TheRealSteve
wrote:

Ok, then let's see if you know what you're talking about. You answer
the question why does only the bayer sensor start to show aliasing
artifacts (in the form of false color) at spatial resolutions lower
than the nyquist spatial resolution of the pixel spacing?

it doesn't. like i said, you're confused.

LOl.. Yes, it does. And I've already given absolute proof that it does
in the form of a test chart showing the nyquist resolution and color
banding occuring before nyquist for a bayer sensor.

*all* sensors will alias before nyquist. there's nothing special about
bayer in that regard. the difference is what the aliasing looks like.

Nope. They won't alias before nyquist. You are still confused on the
basic facts of sampling theory. The only reason the bayer sensor is
the only one of the three types that *does* alias before nyquist is
because it's the only one where the actuall spatial sampling rate
(twice the "true" nyquist frequency) is not the same as the pixel
spacing. And that's due to the fact that the individual color channels
are not sampled at the overall sensor's pixel spacing as they are with
the foveon or 3 sensor system.

The fact that you deny the truth means you are the one who is still
confused.

not in the least.

You're right, not in the least. But much more than the least.

On Sat, 05 May 2012 16:04:32 -0700, nospam
wrote:

In article , TheRealSteve
wrote:

It actually is better because it gives a clue as to why the bayer
sensor has a lower overall resolution than the foveon or 3-sensor
system for the same overall pixel resolution.

Sigma counts each sensel as a pixel.
A 45 MPix Bayer will outperform a "45 MPix" Foveon.
Sorry.

You can be sorry all you want. Just because Sigma uses deceptive
marketing practices doesn't mean anything when it comes to true
spatial resolution. Using Sigma's marketing BS, a 45MP Foveon has
equivelant actual pixel density in terms of spatial resolution as a
15MP bayer sensor and is actually a 15MP sensor in terms of the full
color images it can capture.

so far so good. both have the same spatial resolution.

A 15MP Foveon (using the same definition
of a pixel) compared to a 15MP Bayer? The Foveon will outperform the
Bayer in terms of overall image resolution.

no it won't, and that contradicts what you just said above.

Yes it will, and it does.

no it won't and it doesn't. foveon does not outresolve bayer for the
same pixel count.

Yes it does. Unless you're falling into the marketing trap where Sigma
is trying to convince you that the foveon has 3x the pixels. You're so
confused about everything else, I can believe you believe them.

And it doen not contradict anything

yes it is a contradiction. first you say the spatial resolution is the
same for both bayer and foveon, but then you say the nyquist is
different.

See, you are still totally confused. Yes, the spatial resolution of
the overall pixel locations is the same. The reason that the nyquist
is different for each channel is that the foveon has 3 sensels
(samples all 3 colors) at each pixel location. The bayer only samples
1 color at each pixel location. The sampling of any of the 3 color
channels is much less than the overall pixel spatial resolution, which
is why they are undersampled. When you figure the nyquist rate for the
color channels of the foveon sensor, it's the same as the pixel over
spatial resolution. When you figure the nyquist rate for the color
channels of the bayer sensor, it's less than the overall pixel
resolution. Which is why it's the only one that shows color alias
artifacts at less than the nyquist rate for the overall pixel spatial
resolution.

Untul you realize what actually being sampled, you will remain
confused.

once you
realize that a bayer sensor shows false color alias artifacts at
spatial frequencies lower than nyquist if you figure the nyquist rate
as being the pixel resolution for both of the 15MP sensors.

foveon also shows alias artifacts at spatial frequencies lower than
nyquist.

contrary to fanboi beliefs and ignorant statements by some reviewers,
foveon is subject to the limitations of sampling theory.

You make these statements that try to show you might not be clueless
but they really prove you are,

Foveon is absolutely subject to the limitiations of sampling theory as
the bayer sensor. That's exactly why foveon aliases less for the same
image resolution. Because it's sampling each color at every pixel
location. The bayer is not. That you don't realize that the spatial
resolution of the sampling is what determines the nyquist rate and
that they are different for the bayer vs. the foveon or 3 sensor is
your downfall, and is what's causing your confusion.

The reason
is that the Foveon samples each color channel at the pixel nyquist
rate while the bayer sensor samples each color channel at less than
the pixel nyquist rate.

what matters is the sensor's nyquist, not the individual colour
channels.

And that's why you are confused. You don't understand what is being
sampled and how. Until you do, you will remain confused.

On Sat, 5 May 2012 22:27:42 +0200, Wolfgang Weisselberg
wrote:

TheRealSteve wrote:
On Sat, 5 May 2012 02:37:30 +0200, Wolfgang Weisselberg
TheRealSteve wrote:

It actually is better because it gives a clue as to why the bayer
sensor has a lower overall resolution than the foveon or 3-sensor
system for the same overall pixel resolution.

Sigma counts each sensel as a pixel.
A 45 MPix Bayer will outperform a "45 MPix" Foveon.
Sorry.

You can be sorry all you want. Just because Sigma uses deceptive
marketing practices doesn't mean anything when it comes to true
spatial resolution.

Sigma is the only one offering Foveon sensors and thus they are
the ones defining how Foveon pixels are to be counted. Sorry.

Again, you can be sorry all you want. But it appears you're more
interested in falling into Sigma's marketing hype than understanding
what's really being sampled and how. And that's why you're remain
confused.

But if you want to use their marketing hype to compare, then you have
to use it fully. You can't cherry pick. And using their marketing hype
fully, you would have to compare the 45MP Foveon to a 15MP Bayer since
both are counting sensel locations. Now you see why their marketing
hype is somewhat deceptive. But it does give a simple way of showing
why their resolution is better so maybe it's not hype after all.

In article , TheRealSteve
wrote:

Ok, then let's see if you know what you're talking about. You answer
the question why does only the bayer sensor start to show aliasing
artifacts (in the form of false color) at spatial resolutions lower
than the nyquist spatial resolution of the pixel spacing?

it doesn't. like i said, you're confused.

LOl.. Yes, it does. And I've already given absolute proof that it does
in the form of a test chart showing the nyquist resolution and color
banding occuring before nyquist for a bayer sensor.

*all* sensors will alias before nyquist. there's nothing special about
bayer in that regard. the difference is what the aliasing looks like.

Nope. They won't alias before nyquist.

they absolutely do, as will any sampled system. since you don't
understand this, it's no surprise you get the rest wrong.

You are still confused on the
basic facts of sampling theory.

it's not me who is confused.

The only reason the bayer sensor is
the only one of the three types that *does* alias before nyquist is
because it's the only one where the actuall spatial sampling rate
(twice the "true" nyquist frequency) is not the same as the pixel
spacing. And that's due to the fact that the individual color channels
are not sampled at the overall sensor's pixel spacing as they are with
the foveon or 3 sensor system.

that's not how bayer works. why don't you read some of the many, many
papers on various bayer algorithms. some of them might even be written
so that you can understand them.

In article , TheRealSteve
wrote:

It actually is better because it gives a clue as to why the bayer
sensor has a lower overall resolution than the foveon or 3-sensor
system for the same overall pixel resolution.

Sigma counts each sensel as a pixel.
A 45 MPix Bayer will outperform a "45 MPix" Foveon.
Sorry.

You can be sorry all you want. Just because Sigma uses deceptive
marketing practices doesn't mean anything when it comes to true
spatial resolution.

Sigma is the only one offering Foveon sensors and thus they are
the ones defining how Foveon pixels are to be counted. Sorry.

Again, you can be sorry all you want. But it appears you're more
interested in falling into Sigma's marketing hype than understanding
what's really being sampled and how. And that's why you're remain
confused.

But if you want to use their marketing hype to compare, then you have
to use it fully. You can't cherry pick. And using their marketing hype
fully, you would have to compare the 45MP Foveon to a 15MP Bayer since
both are counting sensel locations. Now you see why their marketing
hype is somewhat deceptive. But it does give a simple way of showing
why their resolution is better so maybe it's not hype after all.

except it's not better.

the photos may *look* better to some people because alias artifacts are
mistaken for real detail, and sigma adds a lot of sharpening and boosts
the contrast too. it's all smoke and mirrors.

meanwhile, the colours are off and there's all sorts of ugly blotching
in the shadows and banding in the highlights.

In article , TheRealSteve
wrote:

It actually is better because it gives a clue as to why the bayer
sensor has a lower overall resolution than the foveon or 3-sensor
system for the same overall pixel resolution.

Sigma counts each sensel as a pixel.
A 45 MPix Bayer will outperform a "45 MPix" Foveon.
Sorry.

You can be sorry all you want. Just because Sigma uses deceptive
marketing practices doesn't mean anything when it comes to true
spatial resolution. Using Sigma's marketing BS, a 45MP Foveon has
equivelant actual pixel density in terms of spatial resolution as a
15MP bayer sensor and is actually a 15MP sensor in terms of the full
color images it can capture.

so far so good. both have the same spatial resolution.

A 15MP Foveon (using the same definition
of a pixel) compared to a 15MP Bayer? The Foveon will outperform the
Bayer in terms of overall image resolution.

no it won't, and that contradicts what you just said above.

Yes it will, and it does.

no it won't and it doesn't. foveon does not outresolve bayer for the
same pixel count.

Yes it does. Unless you're falling into the marketing trap where Sigma
is trying to convince you that the foveon has 3x the pixels. You're so
confused about everything else, I can believe you believe them.

you haven't read my numerous posts on sigma's deceptive and fraudulent
marketing. they are one of the sleaziest and most dishonest companies
around. they flat out lie to their customers, and pixels is just one
small part of it.

the whole sd1 price fiasco was nothing more than a greedy money grab by
the previous ceo. the sd1 merrill is not a new version of the camera
(another lie), it's the same camera as before, just at a more normal
price yet still way high for what it is. contrary to sigma's claim,
they did not 'figure out how to make it cheaper' (yet another lie).
sigma has tens of thousands of unsold sd1 cameras collecting dust so
they had to do something, which was cut the price by $5000 (!) and
added the merrill name.

And it doen not contradict anything

yes it is a contradiction. first you say the spatial resolution is the
same for both bayer and foveon, but then you say the nyquist is
different.

See, you are still totally confused. Yes, the spatial resolution of
the overall pixel locations is the same. The reason that the nyquist
is different for each channel is that the foveon has 3 sensels
(samples all 3 colors) at each pixel location. The bayer only samples
1 color at each pixel location. The sampling of any of the 3 color
channels is much less than the overall pixel spatial resolution, which
is why they are undersampled. When you figure the nyquist rate for the
color channels of the foveon sensor, it's the same as the pixel over
spatial resolution. When you figure the nyquist rate for the color
channels of the bayer sensor, it's less than the overall pixel
resolution. Which is why it's the only one that shows color alias
artifacts at less than the nyquist rate for the overall pixel spatial
resolution.

if you really believe that rubbish, then you haven't a clue. you don't
seem to understand how bayer actually works. you just think you do, and
it's wrong.

Untul you realize what actually being sampled, you will remain
confused.

that's what i say about you.

once you
realize that a bayer sensor shows false color alias artifacts at
spatial frequencies lower than nyquist if you figure the nyquist rate
as being the pixel resolution for both of the 15MP sensors.

foveon also shows alias artifacts at spatial frequencies lower than
nyquist.

contrary to fanboi beliefs and ignorant statements by some reviewers,
foveon is subject to the limitations of sampling theory.

You make these statements that try to show you might not be clueless
but they really prove you are,

actually, it's you who excels at that.

Foveon is absolutely subject to the limitiations of sampling theory as
the bayer sensor.

correct. everything is.

That's exactly why foveon aliases less for the same
image resolution.

it does not, as any test chart will show.

Because it's sampling each color at every pixel
location. The bayer is not. That you don't realize that the spatial
resolution of the sampling is what determines the nyquist rate and
that they are different for the bayer vs. the foveon or 3 sensor is
your downfall, and is what's causing your confusion.

the sampling rate is the *same*, assuming the same number of pixels.

The reason
is that the Foveon samples each color channel at the pixel nyquist
rate while the bayer sensor samples each color channel at less than
the pixel nyquist rate.

what matters is the sensor's nyquist, not the individual colour
channels.

And that's why you are confused. You don't understand what is being
sampled and how. Until you do, you will remain confused.

i understand it quite well.

just how many papers on bayer have you read? and of those, how many
have you actually understood? you obviously didn't read floyd's link of
a couple weeks ago, or if you did, it was well over your head.

TheRealSteve wrote:
On Sat, 5 May 2012 22:27:42 +0200, Wolfgang Weisselberg
TheRealSteve wrote:
On Sat, 5 May 2012 02:37:30 +0200, Wolfgang Weisselberg
TheRealSteve wrote:

It actually is better because it gives a clue as to why the bayer
sensor has a lower overall resolution than the foveon or 3-sensor
system for the same overall pixel resolution.

Sigma counts each sensel as a pixel.
A 45 MPix Bayer will outperform a "45 MPix" Foveon.
Sorry.

You can be sorry all you want. Just because Sigma uses deceptive
marketing practices doesn't mean anything when it comes to true
spatial resolution.

Sigma is the only one offering Foveon sensors and thus they are
the ones defining how Foveon pixels are to be counted. Sorry.

Again, you can be sorry all you want. But it appears you're more
interested in falling into Sigma's marketing hype than understanding
what's really being sampled and how.

I'm simply not falling for *your* *personal* interpretation how
things are supposed to be.

And that's why you're remain confused.

One question: How do you call people who, in discussions, use
words differently from the official and from the common use and
definition and call everyone else confused?


But if you want to use their marketing hype to compare, then you have
to use it fully.

I do.
Bayer -- megapixels as used by the people who make and market
such cameras.
Foveon -- megapixels as used by the people who make and market
such cameras.

You can't cherry pick.

You do it all the time.

And using their marketing hype
fully, you would have to compare the 45MP Foveon to a 15MP Bayer since
both are counting sensel locations.

Why would I have to do that?
Sensels and locations aren't even used in the marketing hype.
Only megapixels.

So, I compare straight: megapixels to megapixels.

But you insist on comparing things you make up for no better reason
that to feel better that you bought a Foveon dog of a camera.

-Wolfgang

In article , Wolfgang
Weisselberg says...
Again, you can be sorry all you want. But it appears you're more
interested in falling into Sigma's marketing hype than understanding
what's really being sampled and how.

I'm simply not falling for *your* *personal* interpretation how
things are supposed to be.

Oh please. Sigma marketing bull**** is Sigma marketing bull****. Period.
--

Alfred Molon
------------------------------
Olympus E-series DSLRs and micro 4/3 forum at
http://tech.groups.yahoo.com/group/MyOlympus/
http://myolympus.org/ photo sharing site

On Mon, 07 May 2012 11:47:48 -0700, nospam
wrote:

In article , TheRealSteve
wrote:

Ok, then let's see if you know what you're talking about. You answer
the question why does only the bayer sensor start to show aliasing
artifacts (in the form of false color) at spatial resolutions lower
than the nyquist spatial resolution of the pixel spacing?

it doesn't. like i said, you're confused.

LOl.. Yes, it does. And I've already given absolute proof that it does
in the form of a test chart showing the nyquist resolution and color
banding occuring before nyquist for a bayer sensor.

*all* sensors will alias before nyquist. there's nothing special about
bayer in that regard. the difference is what the aliasing looks like.

Nope. They won't alias before nyquist.

they absolutely do, as will any sampled system. since you don't
understand this, it's no surprise you get the rest wrong.

You're absolutly wrong as none do. Someone tried to prove they do a
little while ago by showing an example with an o-scope. It was laughed
at as an example because the guy didn't include a reconstruction
filter. So it was getting alias artifacts on reconstruction even
though there were none in the sampled waveform.

The fact that you don't understand basic sampling theory goes a long
way towards explaining why you're so confused on the differences
between the bayer sensor vs. the Foveon and 3 sensor systems.

You are still confused on the
basic facts of sampling theory.

it's not me who is confused.

You've proven that you are several times. The latest is by your
statement above that says you get aliasing in the sampled waveform at
less than the nyquist frequency. That's totally factually wrong and
proves beyond a shadow of a doubt that you don't know what you're
spewing about.

The only reason the bayer sensor is
the only one of the three types that *does* alias before nyquist is
because it's the only one where the actuall spatial sampling rate
(twice the "true" nyquist frequency) is not the same as the pixel
spacing. And that's due to the fact that the individual color channels
are not sampled at the overall sensor's pixel spacing as they are with
the foveon or 3 sensor system.

that's not how bayer works. why don't you read some of the many, many
papers on various bayer algorithms. some of them might even be written
so that you can understand them.

You have no more credibility when you say "that's not how a bayer
works" because you don't know the basics about how anything concerning
digital sampling theory works.