Help understanding PPI and resolution
Tjacobs
negative is about 2200-2400 ppi. What is the point of a scanner that
has 4800 or 5400 ppi? Is it that the resolution will not change at
higher ppi but scanning/adding more ppi will help maintain that
maximum resolution as the enlargement increases? Another question do
slides also have a maximum resolution similar to negative? Thanks for
the education
tom
Rafe B.
wrote:
The issue of "how many megapixels is in a 35 mm frame"
has been debated for years on various USENET forums
and elsewhere. Folks tend to believe what they want on
this issue.
My feeling is that with decent equipment, film and technique
film is "good" for somewhere around 2000-4000 dpi. And
there is no linear relationship here. Here's my best SWAG:
2000 dpi gives you 80% of what's on the film.
2700 dpi gives you 90% of what's on the film.
4000 dpi gives you 95% of what's on the film.
etc.
Like I say, this is a SWAG and the numbers
vary all over the place depending on the film,
the lens, the subject matter, the technique,
and of course the quality of the scan.
rafe b.
http://www.terrapinphoto.com
mkochsch
to calculate the exact DPI of any given frame of emulsion. It really is the
end result that counts anyway right? Yes some people say the "equivalent" is
around 2700 DPI. Having a scanner at double this resolution just means that
you can have two of the scanner's CCD pixels shoot the same relative area of
the frame and give you an average reading of the photoemulsion in theory
giving you slight better definition. It's still not going to allow you to
blow up a 35mm frame much past 16X20 without starting to see the image
degrade into grain. Like I said it's the final product that counts right. If
you're making blow ups of a bride and groom on 35mm and for some reason
aren't using a medium format camera maybe a higher resolution CCD makes
sense if it gives you a slight edge when printing. Also some of these claims
may also be "marketing" babble. Slides have a narrower exposure latitude
(similar to digital) but have more saturation and contrast and aren't
affected by your mini-lab's interpretation of what the final print should
look like. Slides don't work as well in high contrast lighting conditions
ie. landscapes shot at high noon. For beginners slides often seem to scan
easier because the scanner just has to give you what it sees.
~m
"Tjacobs" <thomas...@dva.state.wi.us> wrote in message
news:eaadaf9c.04010...@posting.google.com...
Wayne Fulton
thomas...@dva.state.wi.us says...
No, the issue of "how much detail in film" is rather academic in this
context. The purpose of resolution is for enlargement, and the only
real point is about the size of the enlargement you need.
The real issue is this:
The ratio of (scanning resolution / printing resolution) is the
enlargement factor.
For example (scanning prints):
Scan at 600 dpi, print at 300 dpi, for 600/300 = 2X size (double size)
Scan at 300 dpi, print at 300 dpi, for 300/300 = 1X size (original size)
Scan at 150 dpi, print at 300 dpi, for 150/300 = 1/2X size (half size)
Or same thing for scanning film:
Scan at 2700 dpi, print at 300 dpi, for 2700/300 = 9X size.
9X is about 8x12 inches (about A4 size) from full frame 35 mm.
So if printing at 300 dpi,
2400 dpi will permit 2400/300 = 8x enlargement of the film.
5400 dpi will permit 5400/300 = 18x enlargement.
Other situtions exist too, for example, scan at 2400 dpi to print at 10x
size, which will permit printing that size at 2400/10 = 240 dpi.
Fill in any numbers.
--
Wayne
http://www.scantips.com "A few scanning tips"
mkochsch
picture any better is it? So why bother. At what point do you hit the law of
diminishing returns? Why then not just scan at the theoretical critical dpi
(say 2700 for the sake of arguement) and "push" the print bigger by scaling?
~m
"Wayne Fulton" <nos...@invalid.com> wrote in message
news:MeCdnQTxVdM...@august.net...
Wayne Fulton
says...
>
>
>But scanning a 35mm neg with an electron microscope isn't going to make the
>picture any better is it? So why bother. At what point do you hit the law of
>diminishing returns? Why then not just scan at the theoretical critical dpi
>(say 2700 for the sake of arguement) and "push" the print bigger by scaling?
Because the people with 4000 dpi or 5400 dpi film scanners all disagree with
such notions. <g> Assuming there is a need for this large image, those with
the 4000 dpi or 5400 dpi scanners will assure you that good cases can
certainly get more than is possible at 2700 dpi. Surely your point is not
that they are wrong? Film, exposure situations, processing, and equipment
all vary greatly, and yes, there are cases that cannot benefit much at
extremes. But there are also some that can.
But aside from those capability concerns, then in the practical case the
concern is that one scans to create the size of image that they need for
their purpose. The important practical factor is enlargement size, it
defines what is possible. If one wants to print 18x at 300 dpi, one must
scan at 5400 dpi, otherwise the 5400/300 = 18 math cannot work, and the goal
will not be achieved unless they are willing to settle for different numbers.
Even if they do settle, the enlargement factor still defines what numbers
are possible.
But regarding hypothetical theoretical capability, one doesnt scan at 5400
dpi only because they can, one does because they have the need for that size
image (for example, the 18x at 300 dpi). If one wanted to print 4x5 inches,
one cannot use 5400 dpi for that goal. But if they want to print a 16x20
inch print at 300 dpi from 35 mm film, then they have no other choice.
One does what one must do, and hopefully provides tools that can do it.
Kennedy McEwen
<nos...@invalid.com> writes
>In article <3rpJb.919778$6C4.304339@pd7tw1no>, mkochsc...@THISshaw.ca
>says...
>>
>>
>>But scanning a 35mm neg with an electron microscope isn't going to make the
>>picture any better is it? So why bother. At what point do you hit the law of
>>diminishing returns? Why then not just scan at the theoretical critical dpi
>>(say 2700 for the sake of arguement) and "push" the print bigger by scaling?
>
fact of the matter is that you are into the laws of diminishing return
as soon as you increase the sampling density from 0ppi, because the
contrast reproduced by both the camera lens and the film decrease as the
resolution increases. (In fact, the contrast reproduced by some films
and catadioptric lenses can slightly increase for a short range, but in
the general case, more resolution is only achieved with less contrast.)
It is a bit like life really - you start dying the day you are
conceived!
>Because the people with 4000 dpi or 5400 dpi film scanners all disagree with
>such notions. <g> Assuming there is a need for this large image, those with
>the 4000 dpi or 5400 dpi scanners will assure you that good cases can
>certainly get more than is possible at 2700 dpi. Surely your point is not
>that they are wrong? Film, exposure situations, processing, and equipment
>all vary greatly, and yes, there are cases that cannot benefit much at
>extremes. But there are also some that can.
>
>But aside from those capability concerns, then in the practical case the
>concern is that one scans to create the size of image that they need for
>their purpose. The important practical factor is enlargement size, it
>defines what is possible. If one wants to print 18x at 300 dpi, one must
>scan at 5400 dpi, otherwise the 5400/300 = 18 math cannot work, and the goal
>will not be achieved unless they are willing to settle for different numbers.
>Even if they do settle, the enlargement factor still defines what numbers
>are possible.
>
>But regarding hypothetical theoretical capability, one doesnt scan at 5400
>dpi only because they can, one does because they have the need for that size
>image (for example, the 18x at 300 dpi). If one wanted to print 4x5 inches,
>one cannot use 5400 dpi for that goal. But if they want to print a 16x20
>inch print at 300 dpi from 35 mm film, then they have no other choice.
>
I have to disagree, Wayne, enlargement is only ONE important factor, but
certainly not the only one, and often not even the most significant. I
feel sure that many will consider my disagreement a matter of semantics,
one of definitions of what resolution and sample density mean, but under
this particular topic heading I believe it is important that these are
clarified.
Taking your examples, I know you are aware that there are many ways of
printing an image at 18x enlargement at 300ppi without scanning at
5400ppi, and almost every decent image printing package can achieve it
these days. The 18x enlargement at 300ppi could be achieved by scanning
the original at any sampling density the equipment is capable of and
then resampling the image to 25.5" x 17" @ 300ppi. For example, a 35mm
frame scanned at 2700ppi, which would normally resize directly to 18x
enlargement with only 150ppi, can be resampled by interpolation to
5400ppi, yielding the final image at 300ppi on the page. The question
then arises as to what is the true pixel size on the final print - the
resampled 1/300" or the original sampled pixel from the scanner. In
short, even though it is printed at 300ppi, it is really a 300ppi print
if it has been resampled beforehand - but more on that later. In
addition however, a 25.5"x17" print at 150ppi resampled to 300ppi on the
page will be indistinguishable from a non-resampled 300ppi image at
normal viewing distances - the only difference will be visible on close
inspection, which is not normal viewing conditions.
Similarly, printing at 5x4" (6x4" from a 35mm original) can be achieved
from a 5400ppi scanned original simply by resampling or indeed by
printing at a resolution of around 1275ppi. Now, I know that many will
argue the benefit of printing at such a resolution, and in truth I don't
know of any printer that will support it, although many will support
similar resolutions such as 1200 or 1440ppi. However, not all prints
are made for direct viewing. For example, I file my negatives
interleaved with sheets of "contact" prints, with each frame reproduced
as a 4.75cm wide image on the page. Since these images will normally be
viewed under a magnifying glass while searching, I take full advantage
of the native resolution of all Epson desktop printers and print these
contact pages at 720ppi - the resolution which the Epson print driver
itself uses. I can assure you that the improvement in image content is
significant compared to a 300ppi, or even a 360ppi page. In short, not
all prints need be made at 300ppi - even if that is defined as the
density of the original scanner samples on the final page, ie. ignoring
resampling.
The 300ppi baseline is very arbitrary, and not actually very well suited
to many popular printers - all Epson desktop users should use integer
divisions of the native driver resolution: 720, 360, 240ppi etc.
Otherwise the driver itself will be resampling with a non-integer scale
factor, which can result in artefacts in some cases. Other
manufacturers differ in their native resolution but all inkjet drivers
resample the image to a particular sample density before placing any ink
drops on the page. In the case of Epson desktop printers, irrespective
of the ink dot pitch which can range from 360dpi to over 5760dpi, the
image is resampled to 720ppi by the driver. The Epson professional
range use 360ppi as their native resolution, enabling larger images to
be produced within Photoshop's size limits of 32767 pixels in any axis.
So, in many cases it simply is not possible to talk about a 300ppi image
on the page - the resolution on the page is determined by the printer,
not the scanner or the enlargement.
My point is that deriving the scan resolution simply on the enlargement
factor is not really sensible because it depends on either the final
result or the printer used, and I don't believe it addresses Mr Jacob's
original question (or the further issues raised by MKochsch). I would
even go so far as to say that the fact that you get numbers from this
approach which even approximate the sampling density of popular scanners
to be little more than coincidence (a little more because film and
scanner evolution has been driven by what you can see on the final
print). Enlargement factor is certainly one issue to be considered in
selecting scanner resolution, but it is no more important than a
consideration of the final application of the image and a lot less
important than the consideration of the information that is present on
the original film.
On this last issue, you need to determine how much image is actually
resolved by the film itself and how much can be reproduced by the camera
lens - and the scanner lens too.
The resolution of the lenses is limited by the laws of physics through
diffraction - even a perfect lens has a physically limited resolution,
so it follows that a practical lens, such as that on your camera will be
worse. It just isn't possible to resolve all the way down to the atomic
scale with visible band optics, and you can forget all those paranoid
tales about spaceborne optics that can read your private letters over
your shoulder or your car registration plate too! The resolution limit
of an ideal optic with a circular aperture is given by R = 1 / (W x f#),
where W is the wavelength of the light, f# is the aperture f/# of the
lens and R is the limiting resolution in cy/unit distance. The centre
of the visible spectrum is around 550nm, and is reasonable first order
selection for the wavelength of the light we are using. A similar first
order selection of f/# might be 5.6 - remember, this is for a perfect
lens, your real camera lens will have additional aberrations which will
become significant at faster apertures. So a very good lens could
resolve up to 325cy/mm. Invoking Nyquist's sampling theory, we need two
samples per cycle to be able to unambiguously reproduce the image
generated by such a lens, which means about 650ppmm, or 16500ppi!
That is an incredibly high resolution indeed, and few films could even
come close to resolving that information but, somewhere in Kodak's
inventory you will find an emulsion that is up to that job and more - it
might not come in 35mm format, even on a film backing or respond to
green light, but emulsions with that resolution exist. So, as a first
approximation, that is the answer to the original question - you need to
sample at more than 15000ppi to get every last remnant of information
from the best commercially available cameras on the very best possible
emulsion. And that's where, I believe, MKochsch's question about
diminishing returns comes in.
Considering the lens further, the limiting resolution of a lens is just
that - the resolution at which the lens cannot reproduce any contrast at
all, irrespective of the contrast in the original scene. So it is
pretty meaningless in real terms because, unless the contrast exceeds
the noise of the recording medium, no meaningful information is
recorded. The fraction of contrast produced by a lens plotted against
resolution is called the MTF (Modulation Transfer Function) and, for a
perfect lens it is very close to (but not exactly) a straight line that
reduces from 1 at no resolution up to 0 at the limiting resolution. So,
at half the limiting resolution, a perfect lens will reproduce the
original scene contrast at roughly 50% of that contrast in the image. At
three quarters of the limiting resolution, only 25% of the original
scene contrast will appear in the image. Practical lenses cannot exceed
this line, although they can come very close. The point is that as
resolution increases, the useful image content reduces - and that is
just the lens!
Now consider the film. Forget the esoteric emulsions that Kodak or
other producers manufacture for integrated circuit lithography - what
you are actually interested in is the resolution of the emulsions you
stick in your 35mm camera. Just like the lens, the film has an MTF, and
just like the lens the contrast that the film produces in general
reduces as the resolution increases. The information on most emulsions
is available from the manufacturer data sheet: here is Fuji Velvia 50,
one of the highest resolution colour films available in the 35mm format.
See Figure 20 in the pdf file:
http://www.fujifilm.com/JSP/fuji/epartners/bin/AF3-960E.pdf
One of the things you will see on this curve is that the MTF of the film
is not plotted to its limit - just like the lens, the film has a
limiting resolution (around 200cy/mm in this case) and, just like the
lens, the limiting resolution doesn't really tell you much about the
useful resolution range of the film. What matters is where the MTF
curve crosses the noise floor. For film, the noise is essentially the
graininess of the film, and the amplitude of this noise increases with
resolution.
Image contrast obviously depends on the original scene contrast and also
the MTF of the lens - remember this MTF curve only shows the fraction of
image contrast that the film will reproduce, so actual contrast on the
recorded image (and thus its relative magnitude to grain noise) depends
on the contrast of the image focussed on it. Consequently, some images
of very high contrast material (such as a black and white resolution
test slide, or the silhouette of a mountain with the sun rising behind
it, as produced by a very high quality lens) might extrapolate the MTF
curve to higher resolutions before the noise exceeds the reproduced
contrast.
This crossover resolution is provided in the data sheet for two
different image contrasts in Section 17. For a very high contrast
target of 1000:1, Velvia has a limiting resolution of 160lines/mm (more
correctly line pairs per mm which, in the context of limiting
resolution, is the same as cycles/mm). For a lower contrast image, such
as is typical from the image presented by a good photographic lens on a
typical scene, the limiting resolution is a more modest 80cy/mm.
Comparing the rate of fall off in the MTF chart with this information
gives you some idea of how the noise increases with resolution - between
80 and 160cy/mm 625x higher contrast is required to exceed the noise
floor and, since the MTF itself reduces (by extrapolation, so subject to
normal caveats) from around 20% to 2%, the noise increases by around x6.
This excess noise is significant in scanning, as will be apparent later.
However, invoking Nyquist on the limiting emulsion resolutions, you
require a minimum sampling density 160ppmm (approximately 4000ppi) for
typical photographic subjects up to no more than 320ppmm (approximately
8000ppi) for extremely high contrast subject material. Of course, being
well above 0ppi, between these two ranges is well into the "diminishing
returns" region that MKochsch asked about, so going between them
provides much less real increase in information content than might be
anticipated. The MTF plot of Fuji Velvia shows that the MTF has already
fallen to 30% at around 65cy/mm and reducing by approximately 1/4 for
every 10cy/mm. Even ignoring the noise increase, it is obvious that the
actual information content (the MTF summed up over the resolution range)
increases by only around 15% between 80 and 160cy/mm.
Taking some estimate of grain noise into account, derived from both MTF
and limiting resolution, it would appear that Rafe's SWAG of 95%
information content at 4000ppi is quite close to what would be expected
with the best materials - with less competent materials the 95% point
might be reached with much lower resolutions, perhaps as low as 1600ppi
for some fast films and P&S camera lenses.
However, even this analysis ignores an important aspect of scanning -
grain aliasing. That is the ability of the scanner to sample the noise
at one resolution and reproduce it at another (or rather, the inability
of the scanner to ignore the grain it cannot resolve), and that depends
very much on the engineering design of the scanner itself. For a
typical linear CCD scanner the MTF at the Nyquist limit is around 64%,
and around 10% of the total information content above the scanner's
limiting resolution is reproduced as spurious image content - more
noise. Now, if that is mainly noise, and noise which increases
significantly with resolution as we have seen with Velvia, the scanned
image noise might actually be much more than 10% higher at the limiting
resolution than anticipated directly from the film characteristics.
Clearly then, once grain aliasing is considered, the advantage of
scanning at higher resolutions is even more significant than Rafe's
SWAGs - and if that higher resolution comes with some alias reduction
mechanism then the trade-off is even more significant.
Various approaches to overcoming aliasing in general and grain aliasing
in particular exist, ranging from the Grain Dissolver in the Minolta
flagship scanner to the HyperCCD of the hybrid flatbed/film scanners.
All of these attempt to reduce the amount of image content which is
captured above the sampling density, but do so by reducing the MTF of
the scanner itself. In the case of the Grain Dissolver, this is
achieved by softening the image slightly, whilst the HyperCCD uses a
pixel capture area which exceeds the sample pitch. Without having any
measurements to go on, only the results that I have seen users post
online, I would estimate that the Grain Dissolver in the Minolta scanner
reduces aliased grain noise by 50%, ie. from 10% excess noise down to
5%.
Based on that I would reduce Rafe's SWAGs by 5%-10% in practice without
anti-aliasing filters, and estimate a 50% anti-aliased 5400ppi scanner
(as produced by the current Minolta flagship scanner) capable of
recovering around 95% of the information available on Velvia emulsion.
As you can see, there is a lot more to the original question than just
multiplying an arbitrary print sample density by an enlargement factor,
indeed enlargement only really enters the topic once the captured
resolution exceeds the true output resolution (not an arbitrary figure)
but even that missed the original questions completely.
The original information that Mr Jacobs had, that colour negative film
can only reproduce 2200-2400ppi, was simply false and this clearly led
to the confusion he expressed. I have no information of any film (even
1600ASA emulsions) which have such poor capability, however with some
allowance for camera shake and poorly focussed optics, such figures are
not at all unlikely in reality. However, with good practice and
equipment, film can readily produce much more information than this, so
the need for higher resolution is real - if the intention is to recover
as much image information from the film as possible.
Colour negative and slide film of similar speed and quality have similar
resolution capabilities indeed, if anything, negative film generally has
a higher MTF. However, because the image on a negative film is recorded
at a lower contrast than slide film, the effect of grain means that the
high contrast limiting resolution of the negative film is generally
lower than that of slide film.
Although not definitive, you can confirm this by comparing similar
grades of Fuji film, Provia 100F slide film at:
http://www.fujifilm.com/JSP/fuji/epartners/bin/Provia100f.pdf
and Superia 100 negative film at:
http://www.fuji.fi/documents/13/superia_100_af3025e.pdf
Even though the negative film has a superior MTF (30% at around 95cy/mm
compared to only 55cy/mm on the slide film), the noise is much more
significant and the limiting resolution of the slide film is better, at
140cy/mm compared to only 125cy/mm for the negative emulsion.
--
Kennedy
Yes, Socrates himself is particularly missed;
A lovely little thinker, but a bugger when he's pissed.
Python Philosophers (replace 'nospam' with 'kennedym' when replying)
Wayne Fulton
says...
>a 25.5"x17" print at 150ppi resampled to 300ppi on the
>page will be indistinguishable from a non-resampled 300ppi image at
>normal viewing distances - the only difference will be visible on close
>inspection, which is not normal viewing conditions.
Perhaps, but that is two very different things, only one of which has the
slightest theoretical prayer of showing 300 dpi detail. <g> If one wants to
print 300 dpi at 18x, then one needs 18x300 = 5400 dpi. There isnt much that
is simpler than this. Unless maybe one just wants to argue <g>
>In the case of Epson desktop printers, irrespective
>of the ink dot pitch which can range from 360dpi to over 5760dpi, the
>image is resampled to 720ppi by the driver.
I'm very curious why/how you think this is true? Qimage?
Regardless if true or false, at the paper surface I can easily assure you my
2880 dpi Epson cannot print 360 dpi of pixel detail (even when I can unclog it
<g>). My test was with a small multicolor highly detailed graphic icon
(intentionally not continuous). The purpose is so the pixels are easily
identifiable, so one can see what actually happens, how the pixels are
actually reproduced. The only purpose of printing is of course to reproduce
the pixels. Printing 32x32 pixels at 360 dpi gives a tiny image size of
nearly 0.09x0.09 inches, but no matter, it is exactly the same size as any 32
pixels printed at 360 dpi. An 8x glass really helps to see what happens.
Being generous, this image is only "recognizable" at best. It was
disappointing (crude) at the individual pixel level. Fortunately continuous
images can hide this much better. Lower resolution was superior, the lower
the better actually. So I am uncomfortable with 360, and 720 dpi seems
unimaginable to me, and to think an Epson image is twice better at detail than
a Frontier image is also unimaginable.
My suspicion is that the few people that claim better results must simply be
looking a thin black line in the photo image, like line art. Line art can be
better, but it is not the test of a color photo printer. Better to try to
reproduce that result in the detail of green grass and leaves for example,
anything requiring dithered ink drops of multiple colors, since that is what
inkjet printers do.
Kennedy McEwen
<nos...@invalid.com> writes
>In article <61ELGodr329$Ew...@kennedym.demon.co.uk>, r...@nospam.demon.co.uk
>says...
>
>>a 25.5"x17" print at 150ppi resampled to 300ppi on the
>>page will be indistinguishable from a non-resampled 300ppi image at
>>normal viewing distances - the only difference will be visible on close
>>inspection, which is not normal viewing conditions.
>
>
>Perhaps, but that is two very different things, only one of which has the
>slightest theoretical prayer of showing 300 dpi detail. <g> If one wants to
>print 300 dpi at 18x, then one needs 18x300 = 5400 dpi. There isnt much that
>is simpler than this.
>
different native resolution, so your 300ppi never actually gets onto the
final page. But I think you have missed the point I was making here
Wayne: is a "pixel" a sample from the scanner, a sample from the
software, which may include resampling, or a sample from the printer
driver, which certainly does include resampling if you start with 300ppi
and use an inkjet or reasonable quality laser printer?
Printing with 150cy/in detail on the page requires a MINIMUM of 300ppi
samples on the page or a MINIMUM of 5400ppi samples on the original 35mm
frame. Printing with 300ppi on the page can be achieved in many ways,
as I stated earlier, because 300ppi doesn't say anything at all about
the heritage of those pixels. Not only can they be interpolated, but if
you start at 300ppi and use any Epson inkjet printer they certainly will
be! So, if your 300ppi pixels never appear on the page, what pixels are
you planning to use for your enlargement factor?
In short, your 300ppi baseline is an arbitrary one - it may be relevant
to some printers but is certainly not generic. As such, using it as the
basis for deriving scanning resolution is simply building on sand.
>
>>In the case of Epson desktop printers, irrespective
>>of the ink dot pitch which can range from 360dpi to over 5760dpi, the
>>image is resampled to 720ppi by the driver.
>
>I'm very curious why/how you think this is true? Qimage?
>
You know me a lot better than to suggest anything so crude or
plagiarised! ;-)
Actually, it is the result of careful testing and analysis of the
results as outlined below, augmented by literature searches (Epson have
published many papers and have numerous patents which also contain
information of their methods and ink placement algorithms) and much
discussion with other Epson users on Epson printer mailing lists,
including the sharing of test samples and results from many printers in
the range.
It is now so widely known that I understand some of the Epson
professional printers include the information in their technical manuals
these days, but before that some of us worked it all out for ourselves.
;-)
>Regardless if true or false, at the paper surface I can easily assure you my
>2880 dpi Epson cannot print 360 dpi of pixel detail (even when I can unclog it
><g>). My test was with a small multicolor highly detailed graphic icon
>(intentionally not continuous).
I suggest you try some further tests then Wayne, because that is
certainly not true - I can MEASURE as well as visibly see more detail
than that on my contact print sheets and I have hundreds of those
printed on 720x1440ppi Epson 1270. Not only is the difference between
360 and 720ppi images visible with a lupe, but the difference between
300ppi and 360ppi (due to aliasing) is also clearly visible on many
images. Being derived from 4000ppi originals, these prints contain real
unaliased contrast at their limiting 360cy/in resolutions, they are not
simply fine samples of soft images. All printed on high resolution
glossy paper of course - ink smearing on many papers will reduce the
resolution of the results.
To test the actual resampling density, create a test image of single
pixel lines spaced by a single pixel gap - around 100 such lines are
adequate, with horizontal and vertical patterns. Then print this test
image at different ppi - simple resizing, no resampling, in Photoshop
will suffice. Initially try this at, say 300 - 400ppi in steps of
10ppi, use finer steps around 360ppi, watching the aliased results
carefully. Then run through 720ppi the same way, finally, try 1440ppi.
The aliased images allow you to rapidly converge on the native sampling
density of the driver. At exactly the same sample density as the
driver, the driver will either select all the white or all of the black
pixels and a one pixel shift will change from black to white. Repeat at
numerous colours to prove this is not just a full saturation effect.
Following the proof that the Epson driver resamples at EXACTLY 720ppi
(irrespective of the resolution that ink dots are placed on the page,
which can, and in many cases is, much higher) I suggest looking at some
of the Epson patents and publications on their pseudo random dot
placement algorithm. This uses a process similar to a delta-sigma
analogue to digital conversion in that the error between the average
colour of the ink dots placed per driver sample and the colour of that
sample is redistributed between the 8 neighbouring samples. This
stochastic process means that even though there are a limited number of
tones which can be placed on the paper, there is no actual half tone
cell size - it is completely adaptive depending on the image content.
Consequently, fine colour tonality is achieved by averaging over many
driver samples, but full driver resolution is reproduced if sufficient
contrast is present in the image to exceed the difference between the
minimum tone step in each sample. With six/seven colour printers and
2880/5760dpi placement, that minimum tone step is actually quite small,
in image terms. In effect it is a very simple and efficient resolution
adaptive process.
As for your clogs - switch it off at the front panel, not at the wall
socket! ;-) Seriously, two days ago I fired up an Epson 870 for the
first time in over 10 months. I anticipated some problems so the first
thing I did was a nozzle check print. I carefully examined this with a
10x lupe because I was surprised that it printed perfectly - no nozzle
cleaning, just straight into printing. Much of Epson's reputation is
just user incompetence - no matter how idiot proof they make the
printers, humanity never fails to deliver a better idiot! ;-)
>
>Being generous, this image is only "recognizable" at best. It was
>disappointing (crude) at the individual pixel level. Fortunately continuous
>images can hide this much better. Lower resolution was superior, the lower
>the better actually. So I am uncomfortable with 360, and 720 dpi seems
>unimaginable to me,
>
Well, try it for yourself and see. You can't get much further from a
continuous tone image than the simple line test pattern that I used to
determine the driver native resolution, and once you have found that you
can try anything in between as long as it is printed at an integer
division of the driver native resolution.
>and to think an Epson image is twice better at detail than
>a Frontier image is also unimaginable.
No more unimaginable than the fact that Epson's own consumer desktop
printers produce twice the detail of their much more expensive
professional line-up, but that is certainly the case - prints from their
professional wide format printers simply don't stand up to magnified
viewing, prints from their higher end desktop machines easily do.
I am not saying the image from an Epson inkjet is superior to a Fuji
Frontier print - being continuous tone, the Frontier has no need to
adaptively trade resolution for tonal fidelity as a function of local
image content. The Fuji will deliver full tonal fidelity right up to
its limiting 300ppi resolution all the time. But because it is using a
higher native resolution, a MUCH higher ink dot placement resolution and
a driver which automatically trades tonal fidelity for resolution when
possible, the desktop Epson will outperform it on pure resolution and
detail stakes every single time (provided a sufficiently high resolution
glossy paper, such as Epson Premium Glossy Paper, is used). I can't
comment on other printer manufacturers, but it would not surprise me to
find this to be the general case.
>My suspicion is that the few people that claim better results must simply be
>looking a thin black line in the photo image, like line art. Line art can be
>better, but it is not the test of a color photo printer. Better to try to
>reproduce that result in the detail of green grass and leaves for example,
>anything requiring dithered ink drops of multiple colors, since that is what
>inkjet printers do.
>
I have several files of contact prints that prove otherwise - with every
type of image you like to consider, from fine tonal gradation portraits
to fine detailed structures in still life, architecture and landscapes,
even the odd resolution test pattern - all the very stuff you suggest
will show the difference... and it does!
Try it.
;-)
Wayne Fulton
says...
>
>I suggest you try some further tests then Wayne, because that is
>certainly not true - I can MEASURE as well as visibly see more detail
>than that on my contact print sheets and I have hundreds of those
>printed on 720x1440ppi Epson 1270. Not only is the difference between
>360 and 720ppi images visible with a lupe, but the difference between
>300ppi and 360ppi (due to aliasing) is also clearly visible on many
>images.
Sorry, I've already done my homework, but I cant claim to even see 360, much
less 720 dpi. I think my images are good enough, and I've used the 48MB
PhotoDisk image too, but nothing I try sees it (I can see the lineart, but
that's not my test). I think the little detailed graphics icon (to be able to
compare with what SHOULD be there) is excellent to show why. The printer has
good edge enhancing algorithms, but if the pixels are not reproduced, then
claims are a little harder for me.
I understand about stochastic dithering, but on-average, a 720 dpi pixel's
dimension is obviously only 0.0014 inches... I have to wonder how many CMYK
ink dots will fit in such an area? <g>
You obviously believe it, and power to you, but I am not yet converted.
>As for your clogs - switch it off at the front panel, not at the wall
>socket! ;-) Seriously, two days ago I fired up an Epson 870 for the
>first time in over 10 months. I anticipated some problems so the first
>thing I did was a nozzle check print. I carefully examined this with a
>10x lupe because I was surprised that it printed perfectly - no nozzle
>cleaning, just straight into printing. Much of Epson's reputation is
>just user incompetence - no matter how idiot proof they make the
>printers, humanity never fails to deliver a better idiot! ;-)
I must be that idiot then, because that's also much better luck than I have.
Most of my Epson ink is used to clear clogs. It was OK for awhile, but I'm
tired of the frustratation now, and my next printer will probably be HP again.
Kennedy McEwen
<nos...@invalid.com> writes
>
>I understand about stochastic dithering, but on-average, a 720 dpi pixel's
>dimension is obviously only 0.0014 inches... I have to wonder how many CMYK
>ink dots will fit in such an area? <g>
>
by the printer specification. In the end it really comes down to how
small a drop size they can go to. Some of the latest 7 colour Epson
printers can lay down 32 drops in that space - that in itself defines a
lot of unique colours. Not as many as a continuous tone device, of
course, but certainly plenty sufficient to achieve a meaningful image
representation when combined with the stochastic dither to achieve finer
tonal discrimination across larger dimensions.
Once again though, I suggest that you try your tests again, since the
printers certainly do produce the resolution that they claim when used
with the recommended media. Not only is the basic sampling 720ppi (for
the desktop range) but using that sample density provides a clearly
visible resolution increase over 360ppi - and test images shows print
resolution well beyond the 180cy/in limit of that lower sample density.
I wouldn't claim it goes all the way up to the 360cy/in limit of a
720ppi density, but it is well on the way towards it, and certainly well
beyond the limits of unaided visual acuity.
>
>You obviously believe it, and power to you, but I am not yet converted.
I would like to think you know I am somewhat cynical of manufacturer
specifications and my beliefs are generally based on tests and clear
results, and in this case I can assure you they are. That is precisely
why I use it for printing sheets of small contact print images which
will normally be viewed under magnification.
For normal prints I use a much lower figure for speed and ease of file
handling, usually an integer division of the native driver resolution,
for obvious reasons, which specifically excludes your 300ppi baseline.
mkochsch
transfers....sigh...
Wayne Fulton
says...
>In the end it really comes down to how
>small a drop size they can go to. Some of the latest 7 colour Epson
>printers can lay down 32 drops in that space - that in itself defines a
>lot of unique colours.
This is getting a little wild for me Kennedy.
The carriage/paper motor stepping specs might suggest that, but
physical ink drop size on paper is something entirely different.
The results on paper is all that matters.
It is trivially easy to print a highly detailed multicolor 32x32 pixel icon
scaled to 720 dpi (less than 0.09 inch square), and look at the printed pixels
with a magnifying glass (8x is good). The point is that these non-continuous
pixels are extremely recognizable, so that it becomes an easy case when we
know exactly what it SHOULD look like, pixel by pixel (by simply comparing it
to a very large zoomed copy on the video screen). If it doesnt look like
that (especially if it is hardly recognizable), then we may have a problem
with claims of 720 dpi capability. Reproducing pixels is the only purpose of
printing, pixels are all that exists.
The detailed icon I like for this is to the left of the top paragraph at
http://www.scantips.com/basics1b.html
Inkjet results will appear substantially better at 150 dpi on this test, due
to the greater size of the larger pixels, which allows fitting in a few ink
drops to simulate the color of the pixel.
Kennedy McEwen
<nos...@invalid.com> writes
>In article <GUAUOJI3TS+$Ew...@kennedym.demon.co.uk>, r...@nospam.demon.co.uk
>says...
>
>>In the end it really comes down to how
>>small a drop size they can go to. Some of the latest 7 colour Epson
>>printers can lay down 32 drops in that space - that in itself defines a
>>lot of unique colours.
>
>This is getting a little wild for me Kennedy.
>The carriage/paper motor stepping specs might suggest that, but
>physical ink drop size on paper is something entirely different.
>The results on paper is all that matters.
>
>It is trivially easy to print a highly detailed multicolor 32x32 pixel icon
>scaled to 720 dpi (less than 0.09 inch square), and look at the printed pixels
>with a magnifying glass (8x is good). The point is that these non-continuous
>pixels are extremely recognizable, so that it becomes an easy case when we
>know exactly what it SHOULD look like, pixel by pixel (by simply comparing it
>to a very large zoomed copy on the video screen). If it doesnt look like
>that (especially if it is hardly recognizable), then we may have a problem
>with claims of 720 dpi capability.
So, if it DOES look like the image and all of the key features are
distinguishable when printed, then 720ppi resolution IS achieved.
Once again I suggest, therefore, that you print and examine closely your
own test image, as I have just printed it on an Epson printer of several
years vintage and that is certainly the case. (Your image first
resampled to 32x31 pixels using nearest neighbour pixel selection, then
printed at 720ppi with an Epson 1270 machine on Epson Premium Glossy
Photo Paper.)
By the way, while messing about checking if there might be something
inconsistent between the test image you used and the resolution test
patterns I originally came to these conclusions some time ago from, I
also tried printing out Bart van der Wolf's excellent resampling test
image at different resolutions too. His file is at:
http://www.xs4all.nl/~bvdwolf/main/downloads/Rings.gif
and, although monochrome, is a full grey scale. The susceptibility of
that image to downsampling also proves beyond any doubt that the Epson
driver resamples to exactly 720ppi - and is capable of that resolution.
If you are reading, neat idea to use a zone plate derivation to pull out
spatial harmonic distortion, Bart! ;-)
>Reproducing pixels is the only purpose of
>printing, pixels are all that exists.
>
>The detailed icon I like for this is to the left of the top paragraph at
>http://www.scantips.com/basics1b.html
>
>Inkjet results will appear substantially better at 150 dpi on this test, due
>to the greater size of the larger pixels, which allows fitting in a few ink
>drops to simulate the color of the pixel.
>
This is where I think you are having the problem Wayne. Resolution is
quite independent of tonal fidelity.
Of course printing the same image at 150ppi will produce a better
representation of the colours in the image (although printing at 144ppi
or 180ppi on an Epson will produce even better results - indeed 360ppi
even produces virtually perfect colours on each and every pixel),
however that is quite distinct from the resolution of the image printed
on the page. Can any two dark pixels separated by a single light pixel
be seen in the final image to be distinct from each other or are they
indistinguishable from a continuous blob? That is the classical
definition of resolution and the resolving power of instuments (the
Dawes limit, further refined by Lord Rayleigh) long before computerised
imaging was even conceived.
In your test image, such pixels are clearly distinct even when printed
on a machine which is only capable of depositing two ink dots per 720ppi
pixel, let alone any of the latest machines. So it is indisputable that
the printer actually achieves 720ppi resolution! Note, that the
intervening lighter (or darker) pixel need not even be fully formed to
meet the resolved criteria, simply visible as a distinction between the
two points adjacent to it - however, in the case of your test image, the
gaps are pretty much ideal, in fact almost perfect in one axis, although
slightly smaller in the other.
The colour of the pixels on the page themselves is immaterial to the
actual resolution of the image - exact colour is achieved in an inkjet
(or laser) printer on a macroscopic level, not microscopic one. As
already mentioned, the stochastic dither algorithm will dynamically
trade-off the tonal fidelity to achieve the maximum resolution on the
page. In the case of your test image, where 100% modulation of primary
colours between adjacent pixels is present, the exact tone and colour of
each pixel is sacrificed to maintain the resolution present in the
image. That is the principle of the stochastic dither - the closest
match to the actual pixel colour that is possible by depositing the
available ink drops (in my case any two of 6 colours) is printed and the
error between the colour produced and the required colour is shared
between the 8 nearest neighbouring pixels. As I mentioned above, even
in your test image, near perfect colour is produced in all pixels at
360ppi on my 1270 ie. with a maximum of only 8 dots placed per pixel!
This trade-off between resolution and tonal fidelity is implemented
because your eye is much more sensitive to density of the image detail
than to the colour of the detail itself - a feature that has long been
exploited in colour TV systems where the luminance bandwidth is often 3
to 8x higher than that of the chrominance (depending on the video
standard) and with conventional digital camera sensors (Foveon type
sensors excluded). So the resolution result of the inkjet would stand
even if the image was represented on the page were represented by only
black and white dots and no colour dots at all (which, incidentally, is
not the case).
This is quite distinct from a continuous tone image, such as the Fuji
Frontier or a dye sublimation printer, where the colour fidelity is
maintained to the full image resolution limit. That just happens to be
one of the side effects of the technology, it doesn't have to be that
way for good images to be interpreted by the eye. Indeed, it is the
primary reason that such continuous tone printing systems do not even
require the extreme resolution of discrete tone printers - just enough
so that your unaided eye cannot resolve the individual pixels. However,
even under relatively minor magnification, the pixelation on a Fuji
print is very obvious - not so on a modern photo quality inkjet.
Now there IS a valid question as to which image contains most useful
information - a 300ppi continuous tone image or a 720ppi image composed
of, say 1440x5760dpi dithered ink dots? However that is a much more
complex question to answer, one which involves a great deal of
information theory and access to detailed mathematical models of how the
human eye and brain interpret the information presented to it. Visual
comparison of some of the latest inkjet output with professional
Frontier prints on Crystal paper suggest it is a close run thing, with
the general balance (IMO) being slightly in favour of the inkjets. If
only they could accurately profile the denser output range then it would
be a done deal in the inkjet favour, but they still seem to be much
poorer at fine shadow reproduction.
Wayne Fulton
says...
>(Your image first
>resampled to 32x31 pixels using nearest neighbour pixel selection, then
>printed at 720ppi with an Epson 1270 machine on Epson Premium Glossy
>Photo Paper.)
I didnt follow the resample part... the image is already 34x32 pixels as is
(the small icon-sized icon at the beginning of the first paragraph, next to
the first character of the sentence). I assume the 34 was my crude crop job
from a screen capture, its been awhile. I refered to it, as is, scaled, not
resampled (specifically the original 32 pixels at 0.04 inches) so there would
be absolutely no differences in the pixels. I dont know what you did, but if
you resampled, maybe you found the huge copy of it instead. A resample back
to 32 pixels will contain a few anomalies, wont be as perfect as the small
original in every detail, but I doubt it will influence these results.
>This is where I think you are having the problem Wayne. Resolution is
>quite independent of tonal fidelity.
I am not referring to anything subtle like tonal fidelity. Any kind of
recognizable red or green or yellow pixel is good enough. I am just referring
to reproducing all of the existing pixels. It does seem right that they be in
the ballpark of the right size, bigger than zero. The black size is a bit
overwhelming. I was using a 780 and I'm sure the same paper, wouldnt think
that was a factor either. Again, you are still doing better than I can in
every way, really gotta admire that.
Kennedy McEwen
<nos...@invalid.com> writes
>In article <R0+RmmBOfg+$Ew...@kennedym.demon.co.uk>, r...@nospam.demon.co.uk
>says...
>>(Your image first
>>resampled to 32x31 pixels using nearest neighbour pixel selection, then
>>printed at 720ppi with an Epson 1270 machine on Epson Premium Glossy
>>Photo Paper.)
>
>I didnt follow the resample part... the image is already 34x32 pixels as is
>(the small icon-sized icon at the beginning of the first paragraph, next to
>the first character of the sentence). I assume the 34 was my crude crop job
>from a screen capture, its been awhile. I refered to it, as is, scaled, not
>resampled (specifically the original 32 pixels at 0.04 inches) so there would
>be absolutely no differences in the pixels. I dont know what you did, but if
>you resampled, maybe you found the huge copy of it instead. A resample back
>to 32 pixels will contain a few anomalies, wont be as perfect as the small
>original in every detail, but I doubt it will influence these results.
>
that gives exactly the same image as the original - which is actually a
32x31 pixel image - your original 34x32 crop includes a couple of
background pixels. So we are indeed using the same image.
>>This is where I think you are having the problem Wayne. Resolution is
>>quite independent of tonal fidelity.
>
>I am not referring to anything subtle like tonal fidelity. Any kind of
>recognizable red or green or yellow pixel is good enough. I am just referring
>to reproducing all of the existing pixels. It does seem right that they be in
>the ballpark of the right size, bigger than zero. The black size is a bit
>overwhelming. I was using a 780 and I'm sure the same paper, wouldnt think
>that was a factor either. Again, you are still doing better than I can in
>every way, really gotta admire that.
>
The difference could be down to the build standard of the printer and
the ability to maintain precision as the dots are layed down in
subsequent head sweeps. The inks are deposited in an interleaved manner
as each row of colour nozzles is offset. The 780 is from the lower cost
range but essentially the same technology as the 890/880, the 8" version
of the 1290 which superseded my own printer. I haven't specifically
checked the achieved resolution of that range of printer, however I do
know that the 890/880 is capable of better 720ppi colour output than my
1270, but it is reasonable to assume that something was sacrificed to
get the cost down on the lower range.
Bart van der Wolf
"Kennedy McEwen" <r...@nospam.demon.co.uk> wrote in message
news:R0+RmmBOfg+$Ew...@kennedym.demon.co.uk...
SNIP
> By the way, while messing about checking if there might be something
> inconsistent between the test image you used and the resolution test
> patterns I originally came to these conclusions some time ago from, I
> also tried printing out Bart van der Wolf's excellent resampling test
> image at different resolutions too. His file is at:
> http://www.xs4all.nl/~bvdwolf/main/downloads/Rings.gif
> and, although monochrome, is a full grey scale. The susceptibility of
> that image to downsampling also proves beyond any doubt that the Epson
> driver resamples to exactly 720ppi - and is capable of that resolution.
> If you are reading, neat idea to use a zone plate derivation to pull out
> spatial harmonic distortion, Bart! ;-)
Thanks. It is awefully critical in revealing aliasing effects (because it
has a linear transition from center-low to corner-high spatial
frequencies). Down sampling methods can be easily compared, although the
target is more critical than most everyday scenes (due to absence of large
regularly patterned surfaces in many scenes).
Although some may dismiss artificial targets and prefer real life scenes,
the CGIs have the benefit of repeatability. In that light, it could also
help to print another target of mine, a modification of a Jewell star
image with a sinusoidal grating:
For HP inkjet printers (3.77MB):
http://www.xs4all.nl/~bvdwolf/main/downloads/Jtf60cy-100mm_600ppi.gif
For Epson inkjet printers (5.28MB):
http://www.xs4all.nl/~bvdwolf/main/downloads/Jtf60cy-100mm_720ppi.gif
Either target can be used for determining printout resolution, but if one
wishes to use it for measuring camera (lens+sensor) and/or film+subsequent
scan, I suggest using the right one for 100x100mm output without
resampling.
This target will allow to, amongst others, easily establish the limiting
resolution of a physical reproduction of it (it can also be used for MTF
calculations by tracing concentric modulation at several radii).
If this target is directly printed to glossy paper with the indicated ppi
or higher (no resampling but just by increasing the ppi e.g. to 1200 or
1440 resulting in a 50x50mm output), one can measure the diameter of the
unresolved blur circle on the print. That diameter in millimeters
translates to lp/mm as follows:
(60 / pi) / mm = lp/mm.
The result will not just tell how well a horizontal/vertical pixel
separation is, but it will tell how good the dithering qua;ity can retain
sinusoidal modulation (mimicing real life continuous tone detail rather
than line-art).
SNIP
> Resolution is quite independent of tonal fidelity.
I think that with the introduction of smaller variable pico liter droplets
that partly blend, and better dithering algorithms with higher physical
placement accuracy, it has become possible to produce many shades of a
color in a very small pixel.
HP claims mixing up to 32 ink drops per dot from 6 ink colors (289 levels
of shades of color from a palette of 1.2 million):
http://www.hp.com/cposupport/multifunction/support_doc/bpu02182.html#P85_6259
Of course a larger dithered pixel can be even more accurate after blending
due to lack of visual acuity, but one would be hard pressed to see a
difference.
SNIP
> Now there IS a valid question as to which image contains most useful
> information - a 300ppi continuous tone image or a 720ppi image composed
> of, say 1440x5760dpi dithered ink dots? However that is a much more
> complex question to answer, one which involves a great deal of
> information theory and access to detailed mathematical models of how the
> human eye and brain interpret the information presented to it.
What's more, do we need 'perfectly' accurate information, or is the visual
impression (say after additional tonescaling and sharpening) more
important?
I guess that depends on the application, but my guess is that the
perceptual quality is often more important than accurate pixel colors (due
to error diffusion dithering), especially if one considers variable
viewing conditions of the end results (metamerism and such). Of course
things like reproducing Pantone colors require a different approach.
Bart
Wayne Fulton
says...
>
>Yes I used the large version. A nearest neighbour pixel resample of
>that gives exactly the same image as the original - which is actually a
>32x31 pixel image - your original 34x32 crop includes a couple of
>background pixels. So we are indeed using the same image.
When I do that same resample (Photoshop, nearest neighbor or other modes too),
I get an artifact or two where a new pixel is instead a 4 pixel block, not the
same as the small original. I think that is not unexpected. But the ability
to exceed theoretical perfection has been my problem lately. <g>
Kennedy McEwen
<nos...@invalid.com> writes
>In article <LjSnjdD$rn+$Ew...@kennedym.demon.co.uk>, r...@nospam.demon.co.uk
>says...
>>
>>Yes I used the large version. A nearest neighbour pixel resample of
>>that gives exactly the same image as the original - which is actually a
>>32x31 pixel image - your original 34x32 crop includes a couple of
>>background pixels. So we are indeed using the same image.
>
>When I do that same resample (Photoshop, nearest neighbor or other modes too),
>I get an artifact or two where a new pixel is instead a 4 pixel block, not the
>same as the small original. I think that is not unexpected.
Eh? It seems exceedingly unexpected to me given the image sizes and the
resampling scale! There is something seriously strange with your setup
Wayne.
In light of your comments above, I have just rechecked what happens with
resampling the large image on your site to 32x31 pixels.
Guess what? Exactly the same as I said before - and no 4 pixel
Blocking!
Just for info, to make sure that we are talking about exactly the same
image and process:
The small image you originally meant for me to try is at
http://www.scantips.com/g2/piicon1.gif
This is 34 x 32 pixels (W x H).
The large image that I actually downloaded and tried is at
http://www.scantips.com/g2/piicon2.gif
This is 382 x 368 pixels (W x H).
Using Photoshop 7.01, I resized the large image to 32 x 31 pixels with
"constrain proportions" and "resample image" checked and "Nearest
Neighbour" selected. No blocking occurs in the resulting 32 x 31 pixel
image at all! This isn't really that surprising since 382/32 is
11.9375, and the difference between that and an integer 12 is exactly
1/16th of a pixel. So the final image would need to be *more* than
16pixels from centre to edge in either axis for a mismatch to occur -
which just isn't possible with a 32 x 31 output!
Now it is just possible that you have an older version of PS that scales
from one corner (although. AFAIK, PS has scaled from the centre since at
least v5) then you should get one double thick line near the centre of
the image in both axes - but even that would give only one 4x4 pixel, so
I have no idea why you are getting the strange results you are.
Indeed, I can prove the scaled result is identical to the small image by
pasting a copy of the newly scaled 32x31 image into layer of the
original 34x32 image and setting the layer to difference mode. Guess
what that produces? A completely black result with a single pixel white
border around 3 sides (34-32 = 2 and 32-31 = 1). ie. both images are
identical apart from the excess border, caused by the inaccurate crop of
the original small icon.
I have checked this a couple of times now to be sure that it is 100%
repeatable and my version of PS consistently obeys the mathematics. ;-)
> But the ability
>to exceed theoretical perfection has been my problem lately. <g>
>
Since you are not even achieving a mathematically accurate nearest
neighbour scale, it seems little wonder that you are also having
problems getting a 720ppi scaled image sent to or output by the print
driver. You'll never approach perfection if you start with poor
original material, but you shouldn't assume that what you are getting is
perfection when it falls so far short of what is theoretically possible
without some explanation of where the loss is occurring! :-(
Kennedy McEwen
<bvd...@no.spam> writes
>
>Although some may dismiss artificial targets and prefer real life scenes,
>the CGIs have the benefit of repeatability.
As well as direct measurement of specific parameters. :-)
> In that light, it could also
>help to print another target of mine, a modification of a Jewell star
>image with a sinusoidal grating:
>For HP inkjet printers (3.77MB):
>http://www.xs4all.nl/~bvdwolf/main/downloads/Jtf60cy-100mm_600ppi.gif
>For Epson inkjet printers (5.28MB):
>http://www.xs4all.nl/~bvdwolf/main/downloads/Jtf60cy-100mm_720ppi.gif
>
Thanks, Bart, I have already seen similar test patterns for scanner and
camera measurement. In the case of printing, however, I find this type
of pattern very crude in terms of assessing the actual resolution
because the diameter of the contrast vanishing point circle is extremely
small and difficult to measure accurately. So I prefer wedge
structures, which are essentially segments from the centre of your
pattern expanded in the orthogonal axis. That results in a much more
measurable result.
>
>I think that with the introduction of smaller variable pico liter droplets
>that partly blend, and better dithering algorithms with higher physical
>placement accuracy, it has become possible to produce many shades of a
>color in a very small pixel.
>HP claims mixing up to 32 ink drops per dot from 6 ink colors (289 levels
>of shades of color from a palette of 1.2 million):
>http://www.hp.com/cposupport/multifunction/support_doc/bpu02182.html#P85_6259
>
Similar to the approach to other manufacturers. The latest Epsons place
7 colour (CcMmYKk) drops in 3 sizes from 1.5 to 5pl, so a lot of colour
tones can be achieved in a 720ppi pixel - which in itself is well beyond
the resolution limits of the unaided eye.
>
>What's more, do we need 'perfectly' accurate information, or is the visual
>impression (say after additional tonescaling and sharpening) more
>important?
>I guess that depends on the application, but my guess is that the
>perceptual quality is often more important than accurate pixel colors (due
>to error diffusion dithering), especially if one considers variable
>viewing conditions of the end results (metamerism and such).
Indeed, absolute colour accuracy of the pixels is only important if the
pixels are resolved themselves, and well resolved at that, due to the
differing visual acuity of colour and density. Accurate absolute colour
is improved by the dither algorithm.
Wayne Fulton
says...
>Using Photoshop 7.01, I resized the large image to 32 x 31 pixels with
>"constrain proportions" and "resample image" checked and "Nearest
>Neighbour" selected. No blocking occurs in the resulting 32 x 31 pixel
>image at all!
Thats my problem, I was going (too fast) to 34x32 pixels since the smaller
one was that. That of course doesnt take into account the cropping
differences. However that does not affect trying to print at 720 dpi, as I
never dreamed of resampling the larger one, I was printing the smaller
original.
Bobs
wrote:
This is about exactly what my experience has been, too. There are
diminishing returns when using these very high dpi settings, and huge
image files that are slow to edit etc. It is easy to confuse apparant
sharpness with grain detail, and this is where a good resolution test
target can settle the score. There are some exceptionally
fine-grained films like Velvia, HCC and others that can yield improved
results with hi-res scanning, but only in rare circumstances where
lens resolution and camera movement don't mess things up.
The other side of this is print quality resulting from these scans.
Since pixellization is much more objectionable to the viewer than film
grain is, finer resolution scans can result in more visually
acceptable prints, even if they contain no additional image detail.
Of course various up-sampling methods could be a more sensible (and
quicker) approach to this.
Kennedy McEwen
<nos...@invalid.com> writes
print this at 720ppi then all of the key details are discernible (as
well as pretty close representations of the individual pixel colours),
so 720ppi resolution is certainly produced on the final print, in
agreement with my own tests conducted soon after buying the printer a
few years ago - its good to confirm that the considerable wear and tear
in that time hasn't affected the performance. ;-) Whilst that may be
more than you obtain on your particular printer, it doesn't "exceed
theoretical perfection".
The question then arises as to why your printer is not achieving this,
whilst mine (and those of many people I correspond with) certainly is.
It may be a limitation due to relaxed tolerances on that particular
model, however I would also take a careful look at the print mode you
are using and the printer adjustments. Check that "High Speed Printing"
is off, for the moment, that "Microweave" is on and that "Digital Camera
Correction" or "DCC" is not selected. That should produce the best
image from the printer. Also use the Print Head Alignment utility to
ensure the nozzles are properly aligned to within the 720ppi tolerance,
which will require viewing the test lines with a lupe. Once done
correctly, this should improve the quality of the fine detail,
especially if high speed mode is switched back on.
"DCC" causes the driver to resample the image to 720ppi using linear
interpolation, instead of the faster nearest neighbour resampling. This
reduces the pixelation on images printed from low pixel count sources,
such as low cost digital cameras, and the aliasing amplitude from
sub-optimal (ie non-integer divisions of 720ppi) image sampling
densities. It shouldn't have any effect with a 720ppi source but, since
I haven't specifically tested it on that size, I can't be sure: best
leave it off to be certain it isn't degrading your test. However, it
can allow you to get very acceptable results from quite low resolution
images, where pixelation would normally be very objectionable. Results
are no different from linear resampling to 720ppi in the original
application, just output faster. :-)
I would be interested to know if the suggested settings and adjustments
make an improvement to your print output as viewed under magnification.
In principle, the 780 should yield better results than I get, not worse.
As I said before though, something must have been sacrificed to achieve
the cost reduction and it might have been the unnecessarily tight
tolerance for unaided viewing. The 880/890 that it is derived from
certainly does demonstrate a marginal improvement over my printer,
judging from the test images I have examined.