it's still a 3MP camera

[C.P]

Thanks,Joe, for the explanation!

Now I want to see this sg100 Sigma fanboy guy answer you.Did you notice he disapeared after you proved him wrong??

--
Clay P.

 

[Adam-T]

Yes, the 1D should be capable of something. The 1D is still priced
at $10000 (body only) in Norway, and the 1DS can be yours for
$13000.

the 1D new in the UK costs £3300 = $5280 which is a ripoff considering people have been buying them from Dell for $3200 over in the USA - I paid £1800 ($2880USD) for a refurb which is £200 less than I paid for my D60 10 months ago!.

I suppose it's relative, the difference between the new UK price of a 1D and the Refurb price is the difference between a total ripoff and the biggest bang for buck since the blowout price of the Olympus E10 15 months ago (which was cheaper than a Sony 707 or a Dimage at the time) or the current price for a warrantied used mint D30 from a dealer (about £599)..

--
Please ignore the Typos, I'm the world's worst Typist

The No1 Dedicated 1D forum in the UK -------->

http://www.1dforum.co.uk/php/phpBB2/

 

[sg10]

as your complete ignorance of anything DSP has already been well
demonstrated throughout this thread.

IOWs, you had no idea Foeven counts one R, plus one G, plus one B, as a single pixel, while Canon count the same thing as three.

 

[sg10]

So...what does taste better, Chicken or Beef?

Is there a right answer?

Absolutely, there is a right answer for everyone.

Let me re-phrase....

Is there a universal right answer?

Of course not.....

Of course, everyone has to decide for themselves. My point was that pro reviewers clearly have a vested interest in not upsetting their cash cows for the likes of Foveon-- at least not at this point.

 

[sg10]

Not at all, it would be no different than any 10.2MP Bayer,

A 10MP bayer would have 10 million spatially separated samples to
work with.

The SD9 only has 3.5 million spatial samples. There is
no SW magic than can "unstack" the sensors to make them detect more
spatial information.

Actually, there is, but it involves exploiting CA (chromatic
aberration) and the end result is that you have bands of differing
resolutions.

I use the same type of math in my own Bayer demosaic algorithms,
and will soon be applying it to an SD-9 color interpolation
algorithm.

The way the Foveon sensor works, CA manifests as areas where colors
may be totally out of gamut, and interpolated to very strange
values. I'm working to fix that.

Further, you might be surprised,

I'd be extremely surprised. Over 50 years of science and
engineering knowledge would be obsolete. New textbooks would have
to be written.

Actually, there are ways of doing this that fall within current
texts. Imagine the red, green, and blue are out of registration by
1 full pixel in the corners of the image (not hard to imagine with
many of the wider range zooms). Now, lets make the CA nice and
linear, so that at a/2 the corner radius the misalognment is 1/2
pixel. Then an algorithm that does radial interpolation to resize
the red, green and blue layers back into alighnment will have a
full pizel offset in the corners, which won't increase resolution,
but should eliminate a 1 pixel color fringe. And, in the 1/2 pixel
"ring" there will be an effective doubling of resolution.

you might be able to obtain a stunning image this way by
modeling the falloff rate considering many adjacent pixels.

That buys you exactly zero additional spatial information. Please,
please do some more reading on sampling theory before hypothesizing
magic algorithms.

SG10 definitely needs to do that. You don't have to be out on the
cutting edge like me to realize how many errors are in his
arguments.

You simply haven't read the thread. I'm not saying it would improve image quality. Go back and reread please.

 

[sg10]

Actually, there are ways of doing this that fall within current
texts. Imagine the red, green, and blue are out of registration by
1 full pixel in the corners of the image (not hard to imagine with
many of the wider range zooms). Now, lets make the CA nice and
linear, so that at a/2 the corner radius the misalognment is 1/2
pixel. Then an algorithm that does radial interpolation to resize
the red, green and blue layers back into alighnment will have a
full pizel offset in the corners, which won't increase resolution,
but should eliminate a 1 pixel color fringe. And, in the 1/2 pixel
"ring" there will be an effective doubling of resolution.

Perhaps my imagination is failing me. I can see how this might work
if you make the some assumptions about the target. But otherwise
how do you tell between the CA and real data? And if you can't
reliably detect the difference, aren't we back to arguing over the
difference between resolution and aliasing error?

I certainly was not. You've all completely missed the point. Obviously it makes no sense from an image quality standpoint to unstack the sensors mathematically. Could anything be more obvios? The point is to demonstate 10.2MP equivalency--thats all.

 

[Zone8]

What ever it is, the results are superb and as far as I am concerned, after extensive testing of many others, including the Canon D1s, it beats the lot as far as my needs are concerned.
--
Zone8

 

[Joseph S Wisniewski]

I'll address Dominic's posts in a minute. I remembered an interesting paper where a scientific photographer was trying to use CA to derive spectra in plant photographs, and came across this really interesting bit about an astronomer using CA to derive spectra of stars.

http://home.freeuk.com/m.gavin/chromab.htm

Isn't that just too cool?

I thought about something similar. After reading sg10's statements
I though why not make the 3 layers size a bit different to get a
higher resolution? Isn't that similar?

Unfortunatly, that's not really possible the way a Foveon sensor works. The layers have to be aligned because the only way to sense the well voltage is differentially between layers.

Although a wedge prism in front of the sensor could perform the same function, very easily and cheaply.

Maybe this is part of the
Data that DCRAW ignores? If we have a look on the first versions of
CRW / DCRAW that could decompress X3F the edges of objects were
definitly in different positions in comparison to SPP.

I think that was a bug, demosaicing data that wasn't mosaiced to begin with.

Now they are
in similar positions, maybe this smoothing (bottom-to-top, smooth
right-to-left ...) is the trick and this would also explain why the
output of DCRAW is alot softer than SPP's.

Looking at raw files and SPP files, I've concluded that SPP does sharpen a bit, as well as doing something weird with a diagonal kernal that I haven't unscrambled yet.

Imho this can't be the
sharpening algrithmn alone that makes that difference.

It can.

Does someone see any sense in the above?

Of course.

--
Ciao!

Joe

 

[MdjHtown]

Exactly,

I've read this thread with growing amazement. And I wonder what is really going on here. But then I realized. Say you just forked out a couple of grand for a new digital SLR. Say you bought the 10D or worse forked out more grands for a 1DS. Then this camera comes along that is way cheaper and actually gives them all a run for their money. What will be your first reaction ? .... Exactly. Denial. It can't be. Something is wrong ....

I am surprised they even read this forum. They wouldn't if they weren't intrigued with the quality of the images from this camera.

Cheers y'all.
And now shut up and go shoot some ...
Pictures that is ....
Can't help but think that many a frustrated Canon owner are also NRA members ...
Oops. The jury will disregard that statement !

Cheers Y'all,
Marcel

What ever it is, the results are superb and as far as I am
concerned, after extensive testing of many others, including the
Canon D1s, it beats the lot as far as my needs are concerned.
--
Zone8

 

[Stephen Waits]

IOWs, you had no idea Foeven counts one R, plus one G, plus one B,
as a single pixel, while Canon count the same thing as three.

If that's what you want to believe fine.. the argument is dead, it was when you first started it, and now it's really over.

Now we're just lauging at your stubborn ignorance.. hahahhh.. hoohahaha..

--Steve

 

[Dominic Groß]

Hmm now here we go again after my first reply to this went to dev/null because of Phils lousy windows server. Uhm okay it went to the recycle bin or however this is called in windows.....

http://home.freeuk.com/m.gavin/chromab.htm

Isn't that just too cool?

Yes, very nifty

Maybe this is part of the
Data that DCRAW ignores? If we have a look on the first versions of
CRW / DCRAW that could decompress X3F the edges of objects were
definitly in different positions in comparison to SPP.

I think that was a bug, demosaicing data that wasn't mosaiced to
begin with.

Well but in the first versions with the different edges positions there was no demosaicing and now there is this smoothing. The sharpness was much closer to SPP before this smoothing was added.

Now they are
in similar positions, maybe this smoothing (bottom-to-top, smooth
right-to-left ...) is the trick and this would also explain why the
output of DCRAW is alot softer than SPP's.

Looking at raw files and SPP files, I've concluded that SPP does
sharpen a bit, as well as doing something weird with a diagonal
kernal that I haven't unscrambled yet.

Could this explain the jaggies?

Maybe we should search for someone who is good in disassembling

Dominic

--
http://www.domgross.de
please don't run away because of the cheap design of the first page
ICQ UIN: 289647506

 

[Just Looking]

I've clipped the stuff about advertising claims. As has been
pointed out, the JCII has strict rules on how these things are
defined, and the SD9 can't be marketed in 1/2 the world as anything
other than a 3.4 MP camera. Sigma would look outrageously bad if
they marketed it as different resolutions in different countries.

???

Joe, are you saying there's a new JCIA guideline? or that the "foveon x3 pixel page" has mischaracterized what it says? see here:
http://www.x3f.info/technotes/x3pixel/pixelpage.html

Sigma ads have been saying "10.2 million something", where something has been photo detectors, pixel sensors, and things like that. Have they ever marketed with a "megapixels" number? either 3.4 or 10.2? not that I know of.

j

 

[sg10]

IOWs, you had no idea Foeven counts one R, plus one G, plus one B,
as a single pixel, while Canon count the same thing as three.

If that's what you want to believe fine.. the argument is dead, it
was when you first started it, and now it's really over.

Now we're just lauging at your stubborn ignorance.. hahahhh..
hoohahaha..

Illustrate your maturity level all you want, it won't change the fact that you didn't know 6MP 10D images are interpolated up from only 2MP of full color information. The SD9 could do the same, but chooses not to. Just as Canon could form single pixels by integrating groups of full color sensors and producing a non-interpolated 2MP image, the SD-9 could interpolate up to 10.2MP to produce the same blurry artifact prone image using Bayer algorithms. Sigma should provide the option just to make that point.

 

[sg10]

I've clipped the stuff about advertising claims. As has been
pointed out, the JCII has strict rules on how these things are
defined, and the SD9 can't be marketed in 1/2 the world as anything
other than a 3.4 MP camera. Sigma would look outrageously bad if
they marketed it as different resolutions in different countries.

???
Joe, are you saying there's a new JCIA guideline? or that the
"foveon x3 pixel page" has mischaracterized what it says? see here:
http://www.x3f.info/technotes/x3pixel/pixelpage.html

Sigma ads have been saying "10.2 million something", where
something has been photo detectors, pixel sensors, and things like
that. Have they ever marketed with a "megapixels" number? either
3.4 or 10.2? not that I know of.

Every ad I've seen is quite clear, saying 3.5MP using 10.2M photosites.

j

 

[Dominic Groß]

I've clipped the stuff about advertising claims. As has been
pointed out, the JCII has strict rules on how these things are
defined, and the SD9 can't be marketed in 1/2 the world as anything
other than a 3.4 MP camera. Sigma would look outrageously bad if
they marketed it as different resolutions in different countries.

???
Joe, are you saying there's a new JCIA guideline? or that the
"foveon x3 pixel page" has mischaracterized what it says? see here:
http://www.x3f.info/technotes/x3pixel/pixelpage.html

Sigma ads have been saying "10.2 million something", where
something has been photo detectors, pixel sensors, and things like
that. Have they ever marketed with a "megapixels" number? either
3.4 or 10.2? not that I know of.

Every ad I've seen is quite clear, saying 3.5MP using 10.2M
photosites.

One big shop here is advertising it as 10mp but this is certainly not Sigmas fault. But the censored of Sigma Germany could be censored enough to tell them that it is 10mp....

--
http://www.domgross.de
please don't run away because of the cheap design of the first page
ICQ UIN: 289647506

 

[sg10]

Exactly,

I've read this thread with growing amazement. And I wonder what is
really going on here. But then I realized. Say you just forked out
a couple of grand for a new digital SLR. Say you bought the 10D or
worse forked out more grands for a 1DS. Then this camera comes
along that is way cheaper and actually gives them all a run for
their money. What will be your first reaction ? .... Exactly.
Denial. It can't be. Something is wrong ....
I am surprised they even read this forum. They wouldn't if they
weren't intrigued with the quality of the images from this camera.

Its more than denial, its business. Pro reviewers' livelyhoods depend on playing ball with the influential players. Look at the samples on this very site, the pics make the reality painfully clear, but the conclusions are surprisingly understated. Here is the sample posted...

http://www.pbase.com/imageprocessing

And here is the discussion...

-------------------

"you can see the SD9 image looks sharp and detailed, it's difficult to find any part of the EOS-D60 image which is actually exhibiting more detail than the SD9 and in some places the SD9 manages to pull out detail lost on the EOS-D60. Color balance is similar although it's worth noting the SD9 loses the color at the golden tip of the towers (color clipping problem). You'll also note that I'm not doing any additional processing, you could of course apply an unsharp mask to the EOS-D60 image to improve sharpness (although also introduce new artifacts and amplify noise).

Theoretical reasons why the SD9 images are sharper / exhibit better resolution:

• No interpolation between surrounding pixels, each pixel has its own luminance / color detail
• No low pass (anti-alias filter) to blur detail beyond the Nyquist frequency
• Pixel pitch is larger therefore lens is more capable of delivering adequate resolution "

"At this stage I'm thinking "this is getting interesting". As you can see the SD9 image enlarges quite well, and it shows that it's carrying at least as much resolution / detail as the EOS-D60."

---------------

"Theoretically" better res? Come on, anyone can see it so just say it, there is no contest between the samples posted. The difference in the crop-pair showing redish tones on bluish (worst case for Bayer interpolated images vs non-interpolated images) is nothing short of stunning...



(6th pair from thee top)

Cheers y'all.
And now shut up and go shoot some ...
Pictures that is ....
Can't help but think that many a frustrated Canon owner are also
NRA members ...
Oops. The jury will disregard that statement !

Cheers Y'all,
Marcel

What ever it is, the results are superb and as far as I am
concerned, after extensive testing of many others, including the
Canon D1s, it beats the lot as far as my needs are concerned.
--
Zone8

-

 

[sg10]

I've clipped the stuff about advertising claims. As has been
pointed out, the JCII has strict rules on how these things are
defined, and the SD9 can't be marketed in 1/2 the world as anything
other than a 3.4 MP camera. Sigma would look outrageously bad if
they marketed it as different resolutions in different countries.

???
Joe, are you saying there's a new JCIA guideline? or that the
"foveon x3 pixel page" has mischaracterized what it says? see here:
http://www.x3f.info/technotes/x3pixel/pixelpage.html

Sigma ads have been saying "10.2 million something", where
something has been photo detectors, pixel sensors, and things like
that. Have they ever marketed with a "megapixels" number? either
3.4 or 10.2? not that I know of.

Every ad I've seen is quite clear, saying 3.5MP using 10.2M
photosites.

One big shop here is advertising it as 10mp but this is certainly
not Sigmas fault. But the censored of Sigma Germany could be
censored enough to tell them that it is 10mp....

When I said "every ad" I actually meant every Sigma ad. I have seen shops listing the SD-9 as 10.2MP. Then again, they also advertise the 10D as 6MP, without noting that is an interpolated res, so its really only fair.

 

[Ladehaug]

Illustrate your maturity level all you want, it won't change the
fact that you didn't know 6MP 10D images are interpolated up from
only 2MP of full color information. The SD9 could do the same,
but chooses not to. Just as Canon could form single pixels by
integrating groups of full color sensors and producing a
non-interpolated 2MP image, the SD-9 could interpolate up to 10.2MP
to produce the same blurry artifact prone image using Bayer
algorithms. Sigma should provide the option just to make that
point.

It doesn't work that way. A 10Mp bayer image will in most cases resolve a lot more detail than a 3.4Mp Foveon image. However, the sharpness of a Foveon image resembles what you can get by downsizing a bayer image 50% (to 1/4 of its initial megapixel count). Sharpness and spatial resolution are two different issues.

-
Geir

 

[Ladehaug]

I'm not sure how they got it, but it's close enough to my own
calculations for me to find the number to be comfortable. Mine are
based on the amount of high frequency information lost by an AA
filter, and the amount of luminance information that can be
reconstructed by an optimal Bayer interploation algorithm. The
hypothesized figure of 1.707 was determined by a very ugly circular
integration. It was then checked by analysis of a test set of
images selected at random from our online image library (which
covers everything from machine vision, landscapes, art nudes,
architecture, etc. Suprisingly, you only have to run a few hundred
images before it converges quite nicely.

And yes, we assumed an AA filter in the Foveon case also, although
horizontally and vertically optimal, instead of diagonally as in
the Bayer case, so the spatial frequency limit was 1.414 times
higher.

Thanks, that explains a lot.

Running the Foveon case without an AA filter gives much higher
numbers for distortion. Although the image may appear sharper, this
is false sharpness, similar to "comfort noise".

IMO, the sharpness/aliasing does not disturb the picture at all in most cases. I find that bayer interpolation, on the average, introduces more disturbing artifacts. Of course, the Foveon sensor with an AA filter would be desirable in some cases, but I think it would seriously hamper the sharpness and spatial resolution - something that would not be desirable for a 3.4Mp sensor. I do not know much about AA filters, so I might be wrong. But it I'm sure it would not come cheap to implement and manufacture a high quality AA filter for the Foveon sensor.

-
Geir

 

[sg10]

Illustrate your maturity level all you want, it won't change the
fact that you didn't know 6MP 10D images are interpolated up from
only 2MP of full color information. The SD9 could do the same,
but chooses not to. Just as Canon could form single pixels by
integrating groups of full color sensors and producing a
non-interpolated 2MP image, the SD-9 could interpolate up to 10.2MP
to produce the same blurry artifact prone image using Bayer
algorithms. Sigma should provide the option just to make that
point.

It doesn't work that way. A 10Mp bayer image will in most cases
resolve a lot more detail than a 3.4Mp Foveon image. However, the
sharpness of a Foveon image resembles what you can get by
downsizing a bayer image 50% (to 1/4 of its initial megapixel
count). Sharpness and spatial resolution are two different issues.

I understand exactly what you are saying, but again, Sigma should provide the option just to make that point.