Program to determine optimal reduction without losing significantdetail

Just a curious thought, I wonder if there could be a program to
determine the optimal reduction percentage for a photo without losing
significant detail. Say 95% of original detail. Some photos can go way
way down in size before losing anything. The only method I can think of
is to reduce, then re-enlarge and compare those pixels. That should be
very doable. The test could be performed on a small crop showing the
best detail.

What I'm looking for is an objective way to measure the useful max print
size of a given image, regardless of pixel count. Presumably this would
reflect the strength of the anti-aliasing filter and losses from bayer
interpolation.

--
Paul Furman
www.edgehill.net
www.baynatives.com

all google groups messages filtered due to spam

On Sat, 03 Jan 2009 14:53:19 -0800, Paul Furman
wrote:

Just a curious thought, I wonder if there could be a program to
determine the optimal reduction percentage for a photo without losing
significant detail. Say 95% of original detail. Some photos can go way
way down in size before losing anything. The only method I can think of
is to reduce, then re-enlarge and compare those pixels. That should be
very doable. The test could be performed on a small crop showing the
best detail.

What I'm looking for is an objective way to measure the useful max print
size of a given image, regardless of pixel count. Presumably this would
reflect the strength of the anti-aliasing filter and losses from bayer
interpolation.

It would also have to consider the content of the photo. After all,
no matter what the anti-aliasing filter, bayer interpolation or
original resolution used to take the picture, you can shrink a photo
of a plain white wall pretty far without losing significant detail.

But I'm not sure why you're trying to find the optimal reduction
percentage for a photo if what you're after is an objective way to
measure the useful max print size. I guess I'm just missing
something.

Steve

I don't see how you can disregard pixel-count with regard to detail and
optimum max print size. It may not be the only criteria but it has to be
in there somewhere.

I always strive never to drop below 300ppi for any of my images when I
want to enlarge them, and I avoid using software interpolation to make
up the shortfall -- every pixel must be an original one. And the
'viewing distance' isn't a relevant factor in my book unless you are
printing something like a commercial street-poster.


The final step to aiding preservation of detail is intelligent use of
unsharp mask just prior to printing. Beyond that I see no point in
working harder at it or bringing some other calculations in to the mix.
At 300ppi the maths is simple enough, and the printed results say it
all.

There's nothing quite like the human eye for judging quality of detail
in a print, no software can replace that.

TheRealSteve wrote:
Paul Furman wrote:

Just a curious thought, I wonder if there could be a program to
determine the optimal reduction percentage for a photo without losing
significant detail. Say 95% of original detail. Some photos can go way
way down in size before losing anything. The only method I can think of
is to reduce, then re-enlarge and compare those pixels. That should be
very doable. The test could be performed on a small crop showing the
best detail.

What I'm looking for is an objective way to measure the useful max print
size of a given image, regardless of pixel count.

I might have phrased that awkwardly: I want to reduce the pixel count.


Presumably this would
reflect the strength of the anti-aliasing filter and losses from bayer
interpolation.

It would also have to consider the content of the photo. After all,
no matter what the anti-aliasing filter, bayer interpolation or
original resolution used to take the picture, you can shrink a photo
of a plain white wall pretty far without losing significant detail.

But I'm not sure why you're trying to find the optimal reduction
percentage for a photo if what you're after is an objective way to
measure the useful max print size. I guess I'm just missing
something.

Just a curious thought. I don't know, it would be good for compressing
library archives available online, lots of things. Why save a bigger
file than necessary, especially on screen? Why look at it larger than
needed unless you really want to... some kinds of images look good
enlarged soft but not necessary for most.

I tried a little in photoshop but it's tedious work even if I set up
actions: reduce, re-enlarge, copy, revert to original, paste, set to
difference, add a levels adjustment layer to emphasize the differences.
A program would measure the differences. The photoshop plugin would show
a preview window with the difference layers and an amount slider. Maybe
it wouldn't make much difference, I don't really know, just a thought. I
suppose jpeg does a good enough job of compressing and I guess it does a
better job of preserving fine detail surrounded by white walls but if
that fine detail can be reduced even more without (significant) loss in
(some) photos, that's worth being able to know and control...

--
Paul Furman
www.edgehill.net
www.baynatives.com

all google groups messages filtered due to spam

On Sat, 03 Jan 2009 14:53:19 -0800, Paul Furman wrote:


What I'm looking for is an objective way to measure the useful max print
size of a given image, regardless of pixel count. Presumably this would
reflect the strength of the anti-aliasing filter and losses from bayer
interpolation.

Presumably? ... this will reflect absolutely nothing. All that it will manage to
reflect is that you know nothing about photography, nor cameras, nor
human-values.

What you will discover is what experienced pro-photographers, artists, and
social-scientists have discovered long ago.

Content trumps quality, every time.

It all depends on the content as to what you can increase an image size to. You
must also take into account the perception and values of the intended audience.

A 240x180 image of noisy pixels is highly valuable to a whole field of
scientists if those few pixels are conveying data/information that can be
obtained in no other way. No more resolution nor clarity than that is needed if
the noise-ridden low-resolution image is enough. It can be printed as a two-page
spread in a magazine and all who understand the significance of that noisy
low-resolution image will be in awe over it.

There is no objective test that you can ever do on any image to determine its
empirical maximum print or display size.

Take a well composed 3000x2000 pixel photo of a morning sunrise behind a
mist-hidden waterway and you can effectively print it at billboard size. It will
still be appreciated and admired and nobody will ever notice any pixelation if
properly upsized to smooth the edges of the original pixels. The soft-edged
content of the subject will allow for it. The composition alone being the
appreciable quality.

Take a 640x480 image of an anguished bomb-victim's face, someone who once was
the leader of a powerful country or was a pop-culture icon, and it may be a
full-page image on every newspaper and every magazine-cover in existence. Nobody
will notice the pixelation, nor will they care. The values of society will
determine if that resolution is more than enough--unrelated to any math, camera
resolution, or sensor designs. Those social values always being in flux,
unpredictable, unable to be quantified in any software's algorithms. Anyone
claiming to do so or thinks that it is possible to do so is a blathering fool.

Share a Polaroid photo of a crucifix, star-of-david, or bit of muslim
prayer-carpet properly marinating in a jar of urine and you will turn the world
on its ears.

Content trumps quality, every time.

Maybe if you say it often enough to yourself it will finally sink in. Until
then, keep trying to find tests to tell you how much "resolution" is needed to
convey an image to your audience. Try as you might, until you understand that
simple phrase, you're just mentally masturbating in trying to prove what is a
failed premise. Portraying yourself as just one more of those blathering fools,
as do so many other virtual-photographer resident-trolls in this newsgroup.

Waste your life in trying to finally prove to yourself what experienced pros
have been telling the resident newsgroup trolls for years now. Resolution means
nothing if that resolution doesn't convey anything important to anyone. And then
if it is important information, valuable content for your intended audience,
resolution and clarity still means nothing as long as there's enough to convey
the valuable message contained.

Content trumps quality, every time.

Learn it now, or learn it after a lifetime of failed tests based on
false-beliefs perpetuated by net-parroting usenet-trolls, but that simple fact
will never change and still remain:

Content trumps quality, every time.

In rec.photo.digital.slr-systems Paul Furman wrote:
Just a curious thought, I wonder if there could be a program to
determine the optimal reduction percentage for a photo without losing
significant detail. Say 95% of original detail. Some photos can go way
way down in size before losing anything. The only method I can think of
is to reduce, then re-enlarge and compare those pixels. That should be
very doable. The test could be performed on a small crop showing the
best detail.

What I'm looking for is an objective way to measure the useful max print
size of a given image, regardless of pixel count. Presumably this would
reflect the strength of the anti-aliasing filter and losses from bayer
interpolation.

I use a simple manual method, which is to zoom in to visible pixel
level on a charactertistically highly detailed portion of the image,
and check the effect of applying a small amount of very local
sharpening. If the effect jumps out as a general change of crispness,
then detail exists down to the pixel level. If not, then the image can
be downsized without significant detail loss.

With experience I've become used to how much downsizing a given amount
of softness in the image can take, in terms of 67%, 50%, 33%, 25%, and
other similarly simply vulgar amounts. Using Irfanview it only takes
me several seconds to examine and downsize an image appropriately using
this method.

Obviously this could be automated, but the gain in speed of image
processing wouldn't be worth the programming effort for me, especially
since when I'm being really fussy other considerations influence my
decision and I'd want to do it myself anyway. For example, sometimes I
want to preserve legibility in some very tiny text, and the rest of
the image is naturally softer due to content. I also have a number of
images where the first detail loss on downsizing is being able to
discern individual tiles on a distant roof, and nothing else comes
close. So I have to decide whether being able to spot those tiles
really matters in this image.

So for me a purely quantitative approach, which is all that could be
easily programmed, won't always work.

--
Chris Malcolm

Paul Furman wrote:
Just a curious thought, I wonder if there could be a program to
determine the optimal reduction percentage for a photo without losing
significant detail. Say 95% of original detail. Some photos can go way
way down in size before losing anything. The only method I can think of
is to reduce, then re-enlarge and compare those pixels. That should be
very doable. The test could be performed on a small crop showing the
best detail.

What I'm looking for is an objective way to measure the useful max print
size of a given image, regardless of pixel count. Presumably this would
reflect the strength of the anti-aliasing filter and losses from bayer
interpolation.


Do you mean determining what compression amount to use in the JPEG
compression setting? Or do you mean how much downsampling to allow?

Scott W wrote:
On Jan 3, 12:53 pm, Paul Furman wrote:
Just a curious thought, I wonder if there could be a program to
determine the optimal reduction percentage for a photo without losing
significant detail. Say 95% of original detail. Some photos can go way
way down in size before losing anything. The only method I can think of
is to reduce, then re-enlarge and compare those pixels. That should be
very doable. The test could be performed on a small crop showing the
best detail.

What I'm looking for is an objective way to measure the useful max print
size of a given image, regardless of pixel count. Presumably this would
reflect the strength of the anti-aliasing filter and losses from bayer
interpolation.

A program could do a 2D FFT and look at the highest spatial
frequencies that are in the image. From that it is simple to choose
the lowest sampling frequency that can capture all of the detail that
is in the image.

Hoo boy, that's some abstract stuff. I googled for a while & came up
with what appears to be exactly what you are talking about regarding
digital audio files & chosing the bit depth to use for encoding:
http://ccrma.stanford.edu/~jos/mdft/...ts_Enough.html
It all looks over my head though thanks for the idea.


A complication is that noise in the image will have
high spatial frequencies, so this technique would really only work
well for fairly low noise, i.e. taken at low iso.

True. Noise prevents jpeg from good compression also.

--
Paul Furman
www.edgehill.net
www.baynatives.com

all google groups messages filtered due to spam

Chris Malcolm wrote:
In rec.photo.digital.slr-systems Paul Furman wrote:
Just a curious thought, I wonder if there could be a program to
determine the optimal reduction percentage for a photo without losing
significant detail. Say 95% of original detail. Some photos can go way
way down in size before losing anything. The only method I can think of
is to reduce, then re-enlarge and compare those pixels. That should be
very doable. The test could be performed on a small crop showing the
best detail.

What I'm looking for is an objective way to measure the useful max print
size of a given image, regardless of pixel count. Presumably this would
reflect the strength of the anti-aliasing filter and losses from bayer
interpolation.

I use a simple manual method, which is to zoom in to visible pixel
level on a charactertistically highly detailed portion of the image,
and check the effect of applying a small amount of very local
sharpening. If the effect jumps out as a general change of crispness,
then detail exists down to the pixel level. If not, then the image can
be downsized without significant detail loss.

Yep, this is what I do now though I rarely bother down-sizing. If the
smallest radius sharpening in PS like 0.3 pixels produces benefits,
that's sharp, if it has to be pushed up to 1 or 2, I don't know really
how much it could be reduced... maybe there's a formula as simple as
those numbers so the 2-pixel sharpen job gets reduced .3/2=.15 which is
a whole lot smaller!


With experience I've become used to how much downsizing a given amount
of softness in the image can take, in terms of 67%, 50%, 33%, 25%, and
other similarly simply vulgar amounts. Using Irfanview it only takes
me several seconds to examine and downsize an image appropriately using
this method.

Obviously this could be automated, but the gain in speed of image
processing wouldn't be worth the programming effort for me, especially
since when I'm being really fussy other considerations influence my
decision and I'd want to do it myself anyway.

True, but it might be nice to have an objective starting point. I just
don't think I am very consistent in judging these things, depending on
my mood or whatever I was looking at a moment earlier. Thanks for your
thoughts!


For example, sometimes I
want to preserve legibility in some very tiny text, and the rest of
the image is naturally softer due to content. I also have a number of
images where the first detail loss on downsizing is being able to
discern individual tiles on a distant roof, and nothing else comes
close. So I have to decide whether being able to spot those tiles
really matters in this image.

So for me a purely quantitative approach, which is all that could be
easily programmed, won't always work.



--
Paul Furman
www.edgehill.net
www.baynatives.com

all google groups messages filtered due to spam

Don Stauffer wrote:
Paul Furman wrote:
Just a curious thought, I wonder if there could be a program to
determine the optimal reduction percentage for a photo without losing
significant detail. Say 95% of original detail.

Do you mean determining what compression amount to use in the JPEG
compression setting? Or do you mean how much downsampling to allow?

How much down-sampling to allow.

--
Paul Furman
www.edgehill.net
www.baynatives.com

all google groups messages filtered due to spam