# Tricky shot of an old church

Discussion in 'Photography' started by Scott W, Nov 16, 2005.

1. ### Scott WGuest

Can you show me a real-world waveform where 2x does not work? In the
waveform I showed I can sample at 1/8 the rate as what the data is
shown at, and still get all the info on the waveform.

This is very simple stuff, the highest rate you need to digitize to is
2x the highest frequency that is present in the waveform.

Scott

Scott W, Nov 24, 2005

2. ### Roger N. Clark (change username to rnclark)Guest

Sine X: sample at 0, 180, 360, ...... every 180 degrees, which is
at 2x the frequency.
Result: 0, 0, 0, 0, 0, 0, ....... which means you never see
anything and can't recover anything.

Sampling at 2x the highest frequency is not Nyquist. Nyquist
says at least slightly more than 2x, assuming infinite
sampling. If you don't have infinite sampling, the sampling
rate must go up in order to not induce phase errors.

In the case you posted of the 256 numbers, a damped sine wave,
if you sample every 8th channel and
you reconstruct the sine waves, you do so using an assumption.
There are other equally valid solutions. It is non unique.
Also, it is not a likely response function of a
subject in an imaging system.

Roger

Roger N. Clark (change username to rnclark), Nov 24, 2005

3. ### Roger N. Clark (change username to rnclark)Guest

Scott,
There are two other images at full resolution from the
same 4x5 scan at:
http://www.clarkvision.com/imagedetail/scandetail.html
(3 test areas, the previous grass blades was only one).

But I was hoping people would see "the big picture" before
this. Look at the image again:
http://www.clarkvision.com/imagedetail/resol_exampl1.jpg

The 4x5 image was done at f/45, the 35mm at f/11. Both
lenses are diffraction limited. The focal length of the
4x5 is about 4x that of the 35mm image. Thus, the angular
resolution of the two images should be identical. 4x5
should have no advantage in image detail. Of course every
large format photographer knows that is false, and that
is exactly what the images show. The 35mm image is
very inferior concerning detail. But why? It is due
to the film! The resolution of film for low contrast
subjects is really pretty poor. So the loss in image
detail going to high f/ratios is mitigated somewhat
by the low MTF of film, even high resolution fine grained
films.

This is a big reason why there has been so much debate
over film versus digital: the different MTF responses of the
two media. Digital cameras typically have a higher MTF response
than film, but then drop fast, whereas film keeps going,
but low MTF. But that low MTF response still carries
enough information to give an image the perception of
fine detail.

Then, to tie this back to your original post, the implications
for large digital mosaics like you are producing, have the
potential of showing more detail than large format, even if
the pixel count is less. For example, some of my 4x5 scans
are 12,000 x 16,000 pixels. I would bet an 8,000 x 10,000
pixel digital camera mosaic would look superior (less noise,
more apparent detail) than the 4x5 scan, even if the 4x5 was
done at low f/stop on a perfect lens. That is because
the film is a large part of the limitations to detail
in an image, even 4x5 images.

Roger

Roger N. Clark (change username to rnclark), Nov 24, 2005
4. ### Neil GouldGuest

Well, I understand and agree with Roger's analysis, so see his response.
I'll go further than his comment to add that there will be errors for
*any* phase offset and for almost any waveform sampled at 2x.

Neil

Neil Gould, Nov 24, 2005
5. ### Scott WGuest

The only assumption I need make with my 1/8 th samples is that there
are not frequencies above 1/2 my sampling rate, that is the only
assumption, given that there is only one solution, or can you find
another.

Scott

Scott W, Nov 24, 2005
6. ### Scott WGuest

Ok lets look at the number, if an 8k x 10K digital image looks as good
or better then a scan from a 4 x 5 camera this would say you are
getting about 4 MP/ square inch, this is the number that I tend to use
for film. But this is only if the film shot is done fairly well, good
film and good optics. If you were to shoot that 4 x 5 shot at f/64 it
would not be nearly as good. BTW in your table for comparing digital
to film you have used numbers a bit higher then 4 MP/sq inch for a few
films, I believe close to 7.6 MP / sq inch.

There is no question that 35mm is way limited by the film and to get
away from this limit you need a lot bigger film piece. It is also very
true the if you shot f/16 on 35mm and f/64 on the 4 x 5 the 4 x 5 image
will have a lot more detail, at f/16 on film diffraction hardly an
issue at all, it is almost all the film. But by f/32 we are getting
much softer then at f/16, for any size camera, at least if a very good
film is used, still not enough to greatly reduce the resolution over
what the film MTF is, but clearly softer. But going from f/32 to f/64
is the killer, loosing almost half the resolution. And you know that
you take a big hit when going to f/64, you would only shoot at f/64 if
you had to have the DOF.

This whole discussion started from a statement that diffraction was not
an issue with LF cameras, in many ways it is more of an issue because
to get good DOF requires large f numbers.

Take a look at the sharpest part of a photo that you have taken at f/64
and compare it to the sharpest part of one taken at f/32 and tell me
the one at F/64 is not way softer.

And if we move more in to the extreme, if I were to try and get the
same shot with a 8 x 10 LF camera to get the same shot I would need to
use a lens twice as long and with twice the f number, at f/128 I would
not get much better of a photo shot then shooting the 4 x5 at f/64.

Scott

Scott W, Nov 24, 2005
7. ### Roger N. Clark (change username to rnclark)Guest

Scott,

The original full image for reference, 4x5, 90mm lens, f/64:
http://www.clarkvision.com/galleries/gallery.large_format/web/c072099_L4_01a2-600b.html

30x40 inch, 300 ppi crops:

http://www.clarkvision.com/tmp/sub1-30x40.300ppi.jpg
http://www.clarkvision.com/tmp/sub2-30x40.300ppi.jpg
http://www.clarkvision.com/tmp/sub3-30x40.300ppi.jpg

Print these at 300 ppi and you will see the detail in a
30x40-inxh print. You argued that a 30x40 inch print that the
diffraction spot size about 80 microns in the film plane,
so a 30x40 inch print would produce diffraction spot sizes
of about 640 microns (8x enlargement * 80 microns),
this about 0.64 mm. With this I agree. On a 300 ppi
30x40 inch print that equals about 12 pixels. But the crops
show edge transitions of about 3 pixels in the sharpest areas.
Diffraction spot diameter does not equate to blur circle of all detail.
You get detail up to and beyond about half the radius of the
diffraction disk, consistent with these crops. But the crops show
something much more important than diffraction.

Look at sub1-30x40.300ppi.jpg: the yellow detail in the upper
left is a little soft, the edge of the white flow petal is
in best focus, the purple petal beyond is slightly soft again,
and the green leaves in the background is softer yet,
This illustrates depth of field is a more significant issue
than f/64 diffraction. This is the flower prominent at
bottom center in the full image.

Now look at sub2-30x40.300ppi.jpg: This is a flower that
is to the upper left of sub1, but still slightly below
the center line of the image. It is a little softer due
to the tilt used in combination with the f/64 depth of
field. If I opened up, DOF would be more limiting.

Now look at sub3-30x40.300ppi.jpg: This is a red flower
to the upper right from the sub1 flower, an equal distance
as sub 2 but right, not left. This one is softer yet because
this flower sits in a hole and DOF becomes more of an issue.
Yet edge transitions are still only 3 to 4 pixels.

Grain and scattering in the film is still contributing to softness,
even at f/64. So the loss in detail from f/32 to f/64, for example,
is less than a factor of two.

The point of all this is that real images tell the full story,
whereas one can cite all the numbers one wants, but unless the
model is complete, one might miss important factors.
While diffraction is a factor in large format images, at
f/32, f/45, f/64, etc., it is not the only factor, and should
not be the only consideration in such cases. Film MTF is still
affecting f/64 detail, as is DOF.

Regarding your other numbers, like 4 MP per square inch of film,
this again is putting a number that provides a VERY INCOMPLETE
story of the image. First, megapixels is a function of film
type and speed, so I bristle at people you cite single numbers.
(e.g. film better than digital--yes, no, both are correct-depends
on the situation). See:
http://www.clarkvision.com/imagedetail/film.vs.digital.summary1.html

The above page discusses these points. For example, I rate fuji
velvia at 10 to 16 megapixel equivalent in 35mm, or 7.5 to 12
mpix/sq inch (12 if you want to get the color resolution),
and TMAX 3200 at only about 2.5 megapixel equivalent.

But resolution is not the only thing regarding image quality.
Signal to noise and shape of the MTF response are major factors,
as well as DOF and focus.

The MTF shape and high signal to noise ratio is why digital capture
can get a higher *perceived* image quality than much higher pixel
scanned film. That is described under Apparent Image Quality (AIQ)
on the above page.

So when I said in a digital mosaic that had less pixel count
than scanned 4x5 film, that the digital mosaic could look superior.
I included signal-to-noise, f/ratios and DOF, diffraction,
and different MTF between the two systems. Again, I did not
focus (pun intended) on one number, but considered the whole system.

Roger

Roger N. Clark (change username to rnclark), Nov 25, 2005
8. ### Roger N. Clark (change username to rnclark)Guest

Scott,
I looked at your test data a little more closely. The period of the
highest frequency wave is 21 samples. So your use of every 8 samples
is 2.6 times the frequency, which, by the way is a commonly used
sampling rate in electrical engineering. Even so,
given an unknown function, you do not know the phase, in which case
with only about 10 cycles, you can have different solutions.
As your number of cycles goes down, phase errors can contribute more,
and more solutions are possible.

How about a real imaging example? Below is a table of numbers. The optical
system has a Gaussian response function with a full width at half maximum
of 2 pixels. Sampling is every 1 pixel. Derive the source(s):
position, shape, intensity. The intensities in the table are
8-bit linear (not gamma encoded).

0.0
0.0
1.3
18.0
65.1
80.3
80.0
45.6
7.4
0.3
0.0
0.0
0.0

Roger

Roger N. Clark (change username to rnclark), Nov 26, 2005
9. ### Scott WGuest

With regards to my data, you can not determe the highest frequencies present
by simply counting the periods of the cycles, the envelope of the waveform
effects the bandwidth as well. If you were to look at the fourier transform
of my signal you would see that in fact I am sampling above 2x the max
frequency in the signal.

I did a quick fourier transform on your data and it does not look to be
bandwidth limited to ½ your sample rate, here is a link to what I am seeing.

http://www.sewcon.com/temp/fr.gif

There is clearly a source of energy that is much higher in frequency then
the main signal.

Any mathematician looking at your data would say that it was under sampled,
that there are frequencies at more then 1 / 2 the sampling rate.

Scott

Scott W, Nov 26, 2005
10. ### Scott WGuest

So if these are 300 ppi sample then they would have be scanned at the
equivalent resolution of 2400 ppi on the film?

300 x 40 / 5 = 2400

They are about as soft as many 4000 ppi scans I have seen and softer then
some.

I did print out the samples at 300 ppi, and whereas I am sure it is stunning
as a 30 x 40 inch print if you were to print out a 4 x 6 section it would
look like crap.

If I compare your sample to this one

http://www.terrapinphoto.com/jmdavis/annika2.jpg

they look about the same to my eye in terms of sharpness, but Bret's scan
was done at 5400 ppi, over twice what your scan was done at.

If I take Bret's scan and down sample it to match your it look very sharp,
clearly your scan is not getting even close to all the detail that the film
can produce, and this is what we would expect shooting at f/64. If Bret had
shoot at f/64 his would look just as soft.

If I down sample you image by a factor of two then it starts to look pretty
sharp, but this leaves you would only 27 MP, which is not that much more
then what someone shooting 6 x7 MF with good film can do.

Here is one more way to look at the image, my photo of the church is only 75
MP, I resample and cropped to get a 30 x 40 inch 300 ppi image and pasted a
bit of this next to your crop, seen below.

http://www.sewcon.com/temp/compare.jpg

When printed out there is no comparison in sharpness between the two, but a
sharp clean shoot from a 4 x 5 should be able to produce at least 75 MP, 100
MP would be more like it.

So is my comparison fair? Well no since you had to deal with a lot more DOF,
but that is my point when you go to large f number to get the DOF you suffer
a lot of lost detail. It may be the right trade off to make but you can
pretend that you are not losing the detail.

Scott

Scott W, Nov 26, 2005
11. ### Nicholas O. LindanGuest

As A.Z. once confided to me over a Sachertorte in the Rathaus Park:

"Although it is theoretically feasible to extend this procedure
to more general self-adjoint boundary-value problems associated
with nth order differential operators, in practice this does not
always work since the existence of one single function K(x; t),
which generates all the eigenfunctions of the problem when the
parameter t is replaced by the eigenvalues, is not always
guaranteed."

I think that settles it. Move it to Hilbert space.

Nicholas O. Lindan, Nov 26, 2005
12. ### jetGuest

personally ... it's the side of a church with a steeple ... granted to
you standing there it's beautiful ... since there is little you can do
about parking lots, cars, telephone wires and such ... I would take
that wide angle photograph near the front steps of the church from as
close as you can get (and as wide as you can get) ... it would be
really great if you could get the front of the church and some part of
the steeple in the same photograph ... I would use a 5x7 (110mm) or 6x9
or 6x12 (37mm??) with a wide lens ... to me there may be more story
told from the front of the church than from a side shot of the steeple
and building with parking lot ... just a matter of choice and taste and
that like life is unique amongst all of us ...

jet, Nov 27, 2005
13. ### Roger N. Clark (change username to rnclark)Guest

Nicholas O. Lindan wrote:

Nicholas,
Very cute, but I thought the thread was more like:
Many new large format photographers are enthralled with the
detail in the image and are driven to maximize sharpness.
But many factors influence the overall image, and
sharpness is only one factor.

But Scott has some good points and is illustrating
some new methods, and hopefully both he and I and others reading
are learning (I am). I do see Scott's point that f/64 is
not as sharp, but I already knew that. I don't think other large
format (4x5) photographers use f/64 unless it is needed, but
sometimes it is needed and it is not the factor of 2 loss in
sharpness from f/32 that some might believe. It is not
because film MTF contributes too. I was hoping Scott would
see the light too (perhaps he does).

Roger

Roger N. Clark (change username to rnclark), Nov 27, 2005
14. ### Lorem IpsumGuest

The utter truth is that the picture is not tricky. Not one bit. The photog
is more concerned with technology than making a correct picture.

Lorem Ipsum, Nov 28, 2005
15. ### Scott WGuest

So what do you belive is a good number for sharpness loss when going
from f/32 to f/64, using let's say shooting with Velvia 100, assuming
that the lens is diffraction limited at f/32 and f/64.

And other then limitations of the film what would keep it from being a
loss of 2?

BTW have you seen the scans that Max Perl has done using 35mm Gigabit
film, pretty amazing detail. It really shows how much detail a good
scanner can get if the detail is on the film. I printed out one of his
4000 ppi scans at 300 ppi and the sharpness is amazing.
Here is a link to his scan.
http://www.pbase.com/mxp/image/52664574/original

Scott

Scott W, Nov 28, 2005
16. ### Roger N. Clark (change username to rnclark)Guest

I would say that it is less than a factor of 1.5 loss on
velvia from f/32 to f/64. The loss would be less on higher
speed color films.

The gigabit film is Agfaortho 25. Agfa is now bankrupt.

The guy who does my drum scans showed me a 30x40 (approx)
print of scanned 35mm Kodachrome 25 at 11,000 ppi (the limit
of his scanner). It looked like a 4x5. Unfortunately,
now days velvia is the highest resolution film.

On Max Perl's web site, he also has velvia, and it
looks soft in comparison.

The film industry is going backwards---faster film, lower
resolution.

Roger

Roger N. Clark (change username to rnclark), Nov 28, 2005
17. ### Scott WGuest

There is no doubt that his velvia scans are much softer then the
Gigabit scans, it does show that the scanner is pretty much not
normally the limiting factor, at least if you have a good scanner.

I have notice the trend for faster film for many years now, when I look
at the shelves I don't see anything slower then ISO 400.

As for your 1.5 factor going from f/32 to f/64, I would put it a bit
higher but even a factor of 1.5 is going to loose over half the
information on the film. It is interesting to note that at this loss a
6 x 9 MF camera shooting at f/32 would capture just about the same over
all detail as a 4 x 5 shooting at f/64.

Scott

Scott W, Nov 28, 2005
18. ### Roger N. Clark (change username to rnclark)Guest

Half? 1) the remaining info is at a very low MTF, 2) if it was half,
then there would be a factor of 2 loss, which you agreed it was not that
much. 3) most of the information is at much lower frequencies
(go back to your 2000 ppi test).

Roger

Roger N. Clark (change username to rnclark), Nov 28, 2005
19. ### Scott WGuest

It would be hard to believe you would not loose at least a factor of
1.5 in linear resolution,
which would be a total pixel factor of 2.25.

At some point clearly double the f number would reduce the linear
resolution by a factor of very close to 2. As an extreme example if we
were to go from f/500 to f/1000 we would expect to have almost exactly
half the linear resolution.

So the question is at what f number to we approach this point, clearly
the choice of film has to be taken into account. I believe that at f/32
and higher we are in that range, clearly you thing it would take a
higher f number.

Scott

Scott

Scott W, Nov 28, 2005
20. ### jetGuest

making a "correct" picture is completely personal ... what is correct
to you may not be to others ... personally I find the image less than
thrilling ... but alas it is not my judgement to make ...

jet, Nov 28, 2005