Advice about 12.8 MP Canon Please

Scott W wrote:
David J. Littleboy wrote:

"Scott W" wrote:

There's also the problem of the mathematics of discrete sampling.
The

"capable of recording a distinct line of information for every pixel." bit
sounds suspiciously like perpetual motion and traveling faster than the
speed of light.


Well there is that as well. Assuming that they are talking about one
line being one light or dark line then they can get close to one line
for every pixel. Of course the system still needs a AA filter and for
it to work at all well the contrast for a line pair right at the limit
is going to be close to zero.



That is sort of what I got out of the information. However, it might be
possible in the large 20 µm pixel size, but it might not retain that
capability at smaller sizes.


A good test for just how good an image can look from this kind of
sensor, when not using a AA filter at all, it to take an image and run
a custom filter of a 4 x 4 array of 1s and a scale of 16 next down
sample to 25% using the nearest neighbor sampling method. The 4 x 4
filter is close to the effect that area of the photosite will have,
once down sample by a factor of 4.

Here is a photo I did this to.
http://www.sewcon.com/temp/IMG_8294.jpg

Even if the filter is only 3x3, figure a photosite that is 75% of the
pixel spacing the image looks pretty good.
http://www.sewcon.com/temp/IMG_8294_3x3.jpg

The point is that the physical size of the photosites make a pretty
good AA filter.

Scott


Did you find that physical size varied the anti aliasing performance? In
other words, I would think that larger photosites might have better
performance in that regard.

Ciao!

Gordon Moat
A G Studio
http://www.allgstudio.com

Gordon Moat wrote:
Scott W wrote:
David J. Littleboy wrote:

"Scott W" wrote:

There's also the problem of the mathematics of discrete sampling.
The

"capable of recording a distinct line of information for every pixel."bit
sounds suspiciously like perpetual motion and traveling faster than the
speed of light.


Well there is that as well. Assuming that they are talking about one
line being one light or dark line then they can get close to one line
for every pixel. Of course the system still needs a AA filter and for
it to work at all well the contrast for a line pair right at the limit
is going to be close to zero.



That is sort of what I got out of the information. However, it might be
possible in the large 20 µm pixel size, but it might not retain that
capability at smaller sizes.


A good test for just how good an image can look from this kind of
sensor, when not using a AA filter at all, it to take an image and run
a custom filter of a 4 x 4 array of 1s and a scale of 16 next down
sample to 25% using the nearest neighbor sampling method. The 4 x 4
filter is close to the effect that area of the photosite will have,
once down sample by a factor of 4.

Here is a photo I did this to.
http://www.sewcon.com/temp/IMG_8294.jpg

Even if the filter is only 3x3, figure a photosite that is 75% of the
pixel spacing the image looks pretty good.
http://www.sewcon.com/temp/IMG_8294_3x3.jpg

The point is that the physical size of the photosites make a pretty
good AA filter.

Scott


Did you find that physical size varied the anti aliasing performance? In
other words, I would think that larger photosites might have better
performance in that regard.

A photo site that is the same size as the spacing works out very well,
this is just based on the ratio to the spacing and not the absolute
size of the photo sites. This puts a zero point in the frequency
response at 0.5 line pairs per pixel pixels, this is a good thing.

I should note that there will be a inversion of the phase for
frequencies that go past this, but the amplitude will be low and it
does not seem to cause much of a problem in the images.

Scott

Scott W wrote:
Gordon Moat wrote:

Scott W wrote:

David J. Littleboy wrote:


"Scott W" wrote:

There's also the problem of the mathematics of discrete sampling.
The


"capable of recording a distinct line of information for every pixel." bit
sounds suspiciously like perpetual motion and traveling faster than the
speed of light.


Well there is that as well. Assuming that they are talking about one
line being one light or dark line then they can get close to one line
for every pixel. Of course the system still needs a AA filter and for
it to work at all well the contrast for a line pair right at the limit
is going to be close to zero.



That is sort of what I got out of the information. However, it might be
possible in the large 20 µm pixel size, but it might not retain that
capability at smaller sizes.



A good test for just how good an image can look from this kind of
sensor, when not using a AA filter at all, it to take an image and run
a custom filter of a 4 x 4 array of 1s and a scale of 16 next down
sample to 25% using the nearest neighbor sampling method. The 4 x 4
filter is close to the effect that area of the photosite will have,
once down sample by a factor of 4.

Here is a photo I did this to.
http://www.sewcon.com/temp/IMG_8294.jpg

Even if the filter is only 3x3, figure a photosite that is 75% of the
pixel spacing the image looks pretty good.
http://www.sewcon.com/temp/IMG_8294_3x3.jpg

The point is that the physical size of the photosites make a pretty
good AA filter.

Scott


Did you find that physical size varied the anti aliasing performance? In
other words, I would think that larger photosites might have better
performance in that regard.


A photo site that is the same size as the spacing works out very well,
this is just based on the ratio to the spacing and not the absolute
size of the photo sites. This puts a zero point in the frequency
response at 0.5 line pairs per pixel pixels, this is a good thing.

I should note that there will be a inversion of the phase for
frequencies that go past this, but the amplitude will be low and it
does not seem to cause much of a problem in the images.


In one of the Erwin Puts articles, he put Canon CMOS performance at
about 1.5 pixels per line, based on his testing methods. This is about
consistent with DPReview and a few other sites doing these types of
tests. He seemed to imply that pixel sizes affected this performance,
and the 1.5 number was an average. Wish I could find that reference
again; his site is at http://www.imx.nl in case you are curious.

After looking at tons of test information from various sources, the 1.5
figure seems like a good rule of thumb. I do notice that the Nikon D2X
has (so far) the best outright resolution (not total pixels) of D-SLRs,
and is the smallest of current APS to full frame in regards to cell site
sizes (5.4 µm).

What is missing from these comparisons is hard data and testing of
medium format digital backs. Many are 9 µm cell sites, though a few are
smaller. I read open comments about some lenses lacking resolution
performance when used with some backs . . . . . . just seems crazy when
the numbers might only be 40 lp/mm at best. When I run across tests of
old Russian lenses doing at least that, and someone claiming (Ludicrous
Landscape, et al) Zeiss, Schneider, Rodenstock or Sinar lenses not being
high enough performance . . . well, in the absence of hard data or test
shots, should we really believe such statements?

My guess is that those with medium format digital backs have so much
invested they probably don't want to know actually performance. Why not
just be happy with the money (lot of it) spent, and presume it is "the
best". Of course, another way to look at this is that those who can
afford such gear probably don't have much interest in performance tests;
if the clients like the results, why bother testing.

Ciao!

Gordon Moat
A G Studio
http://www.allgstudio.com

Gordon Moat wrote:

In one of the Erwin Puts articles, he put Canon CMOS performance at
about 1.5 pixels per line, based on his testing methods. This is about
consistent with DPReview and a few other sites doing these types of
tests. He seemed to imply that pixel sizes affected this performance,
and the 1.5 number was an average. Wish I could find that reference
again; his site is at http://www.imx.nl in case you are curious.

After looking at tons of test information from various sources, the 1.5
figure seems like a good rule of thumb. I do notice that the Nikon D2X
has (so far) the best outright resolution (not total pixels) of D-SLRs,
and is the smallest of current APS to full frame in regards to cell site
sizes (5.4 µm).

What is missing from these comparisons is hard data and testing of
medium format digital backs. Many are 9 µm cell sites, though a few are
smaller. I read open comments about some lenses lacking resolution
performance when used with some backs . . . . . . just seems crazy when
the numbers might only be 40 lp/mm at best. When I run across tests of
old Russian lenses doing at least that, and someone claiming (Ludicrous
Landscape, et al) Zeiss, Schneider, Rodenstock or Sinar lenses not being
high enough performance . . . well, in the absence of hard data or test
shots, should we really believe such statements?
I think it is a matter of degree. But over all I would think a good MF
lens could make good use of 9 um pixels spacing. A lot depends on the
lens, a fairly long lens, say 120mm, should not have much problem
producing a very sharp image. A short lens, say 50mm would be much
harder to produce.

My guess is that those with medium format digital backs have so much
invested they probably don't want to know actually performance. Why not
just be happy with the money (lot of it) spent, and presume it is "the
best". Of course, another way to look at this is that those who can
afford such gear probably don't have much interest in performance tests;
if the clients like the results, why bother testing.

I know if I was looking at spending between $20 to $40K on a MF back I
would sure compare it to some much cheaper cameras like the 1Ds MarkII
to get a good idea of just what all that extra money was going to get
me, but that is just me.

Scott

On 20 Apr 2006 17:24:27 -0700, "Scott W" wrote:


I know if I was looking at spending between $20 to $40K on a MF back I
would sure compare it to some much cheaper cameras like the 1Ds MarkII
to get a good idea of just what all that extra money was going to get
me, but that is just me.


I met a photographer a couple of weeks ago who
had tried an MF back (that 39 MP back, IIRC) and
came away disappointed. His interest was with
macro shots, ie., extreme close-ups.

This fellow has two Nikon or Kodak DSLRs, I think
one was the 14 Mpixel version and the other was
a D2X, but don't quote me on that. I think his claim
was that he got better results with the Nikon than
with the 39 MP medium-format back.

In any case, his 2nd effort will be to mount one or
the other of the Nikon bodies onto the film plane
of an Arca-Swiss camera body using a Digitar lens,
and (I guess) using the film-plane shifts to
capture & stitch mulitple images.

His logic -- on why the Hassy back didn't work but
the Nikon/Arca solution would -- was based solely on
the physical frequency (spacing) of sensels.

I don't necessarily buy the logic, but it seems to
be vaguely congruent with the topic at hand.


rafe b
www.terrapinphoto.com

Raphael Bustin wrote:
On 20 Apr 2006 17:24:27 -0700, "Scott W" wrote:


I know if I was looking at spending between $20 to $40K on a MF back I
would sure compare it to some much cheaper cameras like the 1Ds MarkII
to get a good idea of just what all that extra money was going to get
me, but that is just me.


I met a photographer a couple of weeks ago who
had tried an MF back (that 39 MP back, IIRC) and
came away disappointed. His interest was with
macro shots, ie., extreme close-ups.

This fellow has two Nikon or Kodak DSLRs, I think
one was the 14 Mpixel version and the other was
a D2X, but don't quote me on that. I think his claim
was that he got better results with the Nikon than
with the 39 MP medium-format back.

In any case, his 2nd effort will be to mount one or
the other of the Nikon bodies onto the film plane
of an Arca-Swiss camera body using a Digitar lens,
and (I guess) using the film-plane shifts to
capture & stitch mulitple images.
When stitching it helps to move the whole camera and have the lens
only capture a small part of the image in any one image. Close ups are
hard no matter what as getting good resolution and DOF at the same time
is simply not possible.

This is a somewhat close up I did a while back
http://www.sewcon.com/temp/f828.jpg

As close up go it is not all that close, I have some new optics coming
in soon that I hope will let me get much closer.

Scott

"Scott W" wrote:
Gordon Moat wrote:

What is missing from these comparisons is hard data and testing of
medium format digital backs. Many are 9 µm cell sites, though a few are
smaller. I read open comments about some lenses lacking resolution
performance when used with some backs . . . . . . just seems crazy when
the numbers might only be 40 lp/mm at best. When I run across tests of
old Russian lenses doing at least that, and someone claiming (Ludicrous
Landscape, et al) Zeiss, Schneider, Rodenstock or Sinar lenses not being
high enough performance . . . well, in the absence of hard data or test
shots, should we really believe such statements?
I think it is a matter of degree. But over all I would think a good MF
lens could make good use of 9 um pixels spacing. A lot depends on the
lens, a fairly long lens, say 120mm, should not have much problem
producing a very sharp image. A short lens, say 50mm would be much
harder to produce.


Here's the Mamiya 35/3.5 being more than adequate to produce very sharp
images (and Moire, sigh), even with a lot of shift (12mm or so), on a Canon
5D (120 pixels/mm = less than 9 micron pixels). (Caution: big,
overcompressed, oversharpened (I liked what the sharpening was doing to the
textures and forgot to check the tree branches against the sky, oops) file
of ugly test image.)

http://www.pbase.com/davidjl/image/57362779/original

Since this is probably the worst MF lens anyone would be using, and this
would be 24MP in a 36x48mm sensor, I'd say the MF backs will be fine.


My guess is that those with medium format digital backs have so much
invested they probably don't want to know actually performance. Why not
just be happy with the money (lot of it) spent, and presume it is "the
best". Of course, another way to look at this is that those who can
afford such gear probably don't have much interest in performance tests;
if the clients like the results, why bother testing.


I briefly lurked over at the (now defunct) Rob Galbraith MF digital forum,
and they were just figuring out how devastating a problem Moire was...


I know if I was looking at spending between $20 to $40K on a MF back I
would sure compare it to some much cheaper cameras like the 1Ds MarkII
to get a good idea of just what all that extra money was going to get
me, but that is just me.


From what I've seen, the latest round of 22MP and up digital backs are
capturing a lot more detail than the 1Dsmk2. If you need the detail (and can
survive the Moiré), it's either that or scan 4x5...

David J. Littleboy
Tokyo, Japan

David J. Littleboy wrote:
"Scott W" wrote:
I briefly lurked over at the (now defunct) Rob Galbraith MF digital forum,
and they were just figuring out how devastating a problem Moire was...
For objects that don't move a monochrome sensor and color wheel would
work great,
why this is not on the market I don't know. A monochrome sensor should
have very little Morie problems.


Well, no. As above, it wouldn't have _color_ Moiré, but it would have
classical Moiré. And jaggies. You really can't present a periodic arraywith
frequencies above Nyquist and not see problems.

The photo sites can act well as a AA filter. Whereas there is a small
leak of higher frequency energy past the Nyquist limit it is pretty
small.

This is a pretty close simulation as to what the image should look like
http://www.sewcon.com/temp/IMG_8294_3x3.jpg

You should also note that to get the real problem with Moiré you need
a frequency that is much higher then the Nyquist limit. For the
classic Moiré pattern to show up you need to fold the high frequencies
down close to DC, and this take close to twice the Nyquist limit.

Scott


Scott

"Mark Conrad" wrote in message
...
In article , Neil
Gould wrote:

Well... the wrong newsgroup, for starters! ;-)

You may wish to try one of the digital camera groups for information.


Would, if I could find such a NG.

Did searching trying "photo", "digital", "cameras", but no joy. Closest
I came was a lot of Microsoft NGs, which does me no good a a Mac user.

usenet: rec.photo.digital.slr-systems

Gordon Moat wrote:

if the clients like the results, why bother testing.


Why indeed!

;-)