Image enlargement software

On 2014-10-17 23:00:42 +0000, Eric Stevens said:

If you ask the printer to print twice the size it's almost certainly
going to do it by printing twice as many pixels.

"Almost" certainly? How about "possibly" - after all we really don't
know and it depends on the individual printer, and possibly driver
version as well.

What it will have to
do is create a new pixel between each of the original ones. If the
values of the originals were A and B, it is almost certain that the
values assigned to the new one will be (A + B)/2. That might be OK but
if A and B deliniated an edge they would be markedly different and the
new intermediate pixel would soften that edge. That almost certainly
is not what you want.

It might be that the people who do driver development have been
instructed to do just this. It might also be a little bit more
complicated than that - interpolation with edge detection and
sharpening has been a thing in image processing software since the
1980s. The technique may be applied, or not. Again the only way to
know is to test it for yourself. This isn't like an argument about
angels dancing on the head of a pin, this is something that a person
who has an inkjet printer can actually test and see.

The matter is further complicated by the fact that (except at the
edges) each pixel has 8 neighbours in the first surrounding layer and
16 neighbours in the layer surrounding that. Interpolating the values
of new pixels is both a mathematical nightmare and computationally
heavy. It's better to leave the up-scaling/rescanning of your image to
purpose designed software and only rely on the printer driver to
translate the resulting pixel values to the right pattern of droplets
on the paper.

But how do you know that this all the printer is doing? Without
checking, you don't. It might be that image scaling algorithms inside
the printer have been designed with some attention and care, and that
they can do an excellent job. Who knows? Only the people who check
and see.

On 2014-10-17 03:01:34 +0000, Eric Stevens said:

On Thu, 16 Oct 2014 19:13:34 -0700, Oregonian Haruspex
wrote:

On 2014-10-16 00:12:27 +0000, Eric Stevens said:

On Wed, 15 Oct 2014 16:23:11 -0700, Oregonian Haruspex
wrote:

On 2014-10-15 20:54:54 +0000, Eric Stevens said:

So you are saying that currently the best way to enlarge an image with
interpolation is getting someone else to do it. Now, say that 'someone
else' approached you and asked you what is currently the best way to
enlarge an image with interpolation, what would your answer be?

Listen, before the ink shoots out of the nozzles the printer is going
to be doing its own transforms in both the geometric and color domains
anyway, and it's just possible that the fellas who designed its
mechanism and firmware knew what they were doing, no? I mean it's also
possible (but not likely) that things might not turn out that great,
but you can't just take it as a given that in-printer interpolation is
worse than doing it yourself, because doing enlargement yourself in
software is no guarantee that this combined with the printer's own
transforms that it ALWAYS applies as a natural part of the printing
process won't make things worse in the end.

Each printer has its own character and you have to figure out the best
workflow afresh if you change printers.

Hopefully, if you have set things up correctly yourself, the printer
has to do the minimum of adjustment and if your color management is
correct you should end up with what you saw on the screen.

Well it's not like the printer has to do a minimum or maximum of
adjustment, or even that you know what's going on. My only points are
that 1) doing enlargement yourself is not always going to be better and
2) the only way to discover this for any given printer is
experimentation.

There is some very good software out there for rescanning images and I
expect you would have to have a very (very) good printer to do better
than the better available software.

You "would expect" but who knows? Why not try it and find out? After
all, if you are looking to know for sure this is the only way.

On Tue, 21 Oct 2014 16:13:26 -0700, Oregonian Haruspex
wrote:

On 2014-10-17 03:01:34 +0000, Eric Stevens said:

On Thu, 16 Oct 2014 19:13:34 -0700, Oregonian Haruspex
wrote:

On 2014-10-16 00:12:27 +0000, Eric Stevens said:

On Wed, 15 Oct 2014 16:23:11 -0700, Oregonian Haruspex
wrote:

On 2014-10-15 20:54:54 +0000, Eric Stevens said:

So you are saying that currently the best way to enlarge an image with
interpolation is getting someone else to do it. Now, say that 'someone
else' approached you and asked you what is currently the best way to
enlarge an image with interpolation, what would your answer be?

Listen, before the ink shoots out of the nozzles the printer is going
to be doing its own transforms in both the geometric and color domains
anyway, and it's just possible that the fellas who designed its
mechanism and firmware knew what they were doing, no? I mean it's also
possible (but not likely) that things might not turn out that great,
but you can't just take it as a given that in-printer interpolation is
worse than doing it yourself, because doing enlargement yourself in
software is no guarantee that this combined with the printer's own
transforms that it ALWAYS applies as a natural part of the printing
process won't make things worse in the end.

Each printer has its own character and you have to figure out the best
workflow afresh if you change printers.

Hopefully, if you have set things up correctly yourself, the printer
has to do the minimum of adjustment and if your color management is
correct you should end up with what you saw on the screen.

Well it's not like the printer has to do a minimum or maximum of
adjustment, or even that you know what's going on. My only points are
that 1) doing enlargement yourself is not always going to be better and
2) the only way to discover this for any given printer is
experimentation.

There is some very good software out there for rescanning images and I
expect you would have to have a very (very) good printer to do better
than the better available software.

You "would expect" but who knows? Why not try it and find out? After
all, if you are looking to know for sure this is the only way.

I have a *stack* of A2 prints with which I have been experimenting to
answer this question. I gave up leaving it to the printer years ago.
In any case, the question has been answered by others. I don't know of
any book dealing with the production of high quality prints which
recommends just leaving it to the printer.
--

Regards,

Eric Stevens

On Tue, 21 Oct 2014 16:09:18 -0700, Oregonian Haruspex
wrote:

On 2014-10-17 23:00:42 +0000, Eric Stevens said:

If you ask the printer to print twice the size it's almost certainly
going to do it by printing twice as many pixels.

"Almost" certainly? How about "possibly" - after all we really don't
know and it depends on the individual printer, and possibly driver
version as well.

What it will have to
do is create a new pixel between each of the original ones. If the
values of the originals were A and B, it is almost certain that the
values assigned to the new one will be (A + B)/2. That might be OK but
if A and B deliniated an edge they would be markedly different and the
new intermediate pixel would soften that edge. That almost certainly
is not what you want.

It might be that the people who do driver development have been
instructed to do just this. It might also be a little bit more
complicated than that - interpolation with edge detection and
sharpening has been a thing in image processing software since the
1980s. The technique may be applied, or not. Again the only way to
know is to test it for yourself. This isn't like an argument about
angels dancing on the head of a pin, this is something that a person
who has an inkjet printer can actually test and see.

The matter is further complicated by the fact that (except at the
edges) each pixel has 8 neighbours in the first surrounding layer and
16 neighbours in the layer surrounding that. Interpolating the values
of new pixels is both a mathematical nightmare and computationally
heavy. It's better to leave the up-scaling/rescanning of your image to
purpose designed software and only rely on the printer driver to
translate the resulting pixel values to the right pattern of droplets
on the paper.

But how do you know that this all the printer is doing? Without
checking, you don't. It might be that image scaling algorithms inside
the printer have been designed with some attention and care, and that
they can do an excellent job. Who knows? Only the people who check
and see.

It might. But then it might not.

I suggest you do a bit of research. There is a lot of information on
the web.
--

Regards,

Eric Stevens

On 2014-10-22 03:58:55 +0000, Eric Stevens said:

On Tue, 21 Oct 2014 16:13:26 -0700, Oregonian Haruspex
wrote:

On 2014-10-17 03:01:34 +0000, Eric Stevens said:

On Thu, 16 Oct 2014 19:13:34 -0700, Oregonian Haruspex
wrote:

On 2014-10-16 00:12:27 +0000, Eric Stevens said:

On Wed, 15 Oct 2014 16:23:11 -0700, Oregonian Haruspex
wrote:

On 2014-10-15 20:54:54 +0000, Eric Stevens said:

So you are saying that currently the best way to enlarge an image with
interpolation is getting someone else to do it. Now, say that 'someone
else' approached you and asked you what is currently the best way to
enlarge an image with interpolation, what would your answer be?

Listen, before the ink shoots out of the nozzles the printer is going
to be doing its own transforms in both the geometric and color domains
anyway, and it's just possible that the fellas who designed its
mechanism and firmware knew what they were doing, no? I mean it's also
possible (but not likely) that things might not turn out that great,
but you can't just take it as a given that in-printer interpolation is
worse than doing it yourself, because doing enlargement yourself in
software is no guarantee that this combined with the printer's own
transforms that it ALWAYS applies as a natural part of the printing
process won't make things worse in the end.

Each printer has its own character and you have to figure out the best
workflow afresh if you change printers.

Hopefully, if you have set things up correctly yourself, the printer
has to do the minimum of adjustment and if your color management is
correct you should end up with what you saw on the screen.

Well it's not like the printer has to do a minimum or maximum of
adjustment, or even that you know what's going on. My only points are
that 1) doing enlargement yourself is not always going to be better and
2) the only way to discover this for any given printer is
experimentation.

There is some very good software out there for rescanning images and I
expect you would have to have a very (very) good printer to do better
than the better available software.

You "would expect" but who knows? Why not try it and find out? After
all, if you are looking to know for sure this is the only way.

I have a *stack* of A2 prints with which I have been experimenting to
answer this question. I gave up leaving it to the printer years ago.
In any case, the question has been answered by others. I don't know of
any book dealing with the production of high quality prints which
recommends just leaving it to the printer.

What sort of methodology have you been using in your experiments? It'd
be interesting to hear about your test setup and what your metrics are
that would lead you to conclude that one print is better than another.

On 2014-10-22 04:01:57 +0000, Eric Stevens said:

On Tue, 21 Oct 2014 16:09:18 -0700, Oregonian Haruspex
wrote:

On 2014-10-17 23:00:42 +0000, Eric Stevens said:

If you ask the printer to print twice the size it's almost certainly
going to do it by printing twice as many pixels.

"Almost" certainly? How about "possibly" - after all we really don't
know and it depends on the individual printer, and possibly driver
version as well.

What it will have to
do is create a new pixel between each of the original ones. If the
values of the originals were A and B, it is almost certain that the
values assigned to the new one will be (A + B)/2. That might be OK but
if A and B deliniated an edge they would be markedly different and the
new intermediate pixel would soften that edge. That almost certainly
is not what you want.

It might be that the people who do driver development have been
instructed to do just this. It might also be a little bit more
complicated than that - interpolation with edge detection and
sharpening has been a thing in image processing software since the
1980s. The technique may be applied, or not. Again the only way to
know is to test it for yourself. This isn't like an argument about
angels dancing on the head of a pin, this is something that a person
who has an inkjet printer can actually test and see.

The matter is further complicated by the fact that (except at the
edges) each pixel has 8 neighbours in the first surrounding layer and
16 neighbours in the layer surrounding that. Interpolating the values
of new pixels is both a mathematical nightmare and computationally
heavy. It's better to leave the up-scaling/rescanning of your image to
purpose designed software and only rely on the printer driver to
translate the resulting pixel values to the right pattern of droplets
on the paper.

But how do you know that this all the printer is doing? Without
checking, you don't. It might be that image scaling algorithms inside
the printer have been designed with some attention and care, and that
they can do an excellent job. Who knows? Only the people who check
and see.

It might. But then it might not.

I suggest you do a bit of research. There is a lot of information on
the web.

No information on the web can be sufficiently extrapolated to the point
where a person can be sure if their own enlargement voodoo is better
than the in-printer voodoo, unless the person on the web has the same
printer (and driver version!), has given sufficient information about
their own tests so that a person can replicate the results, and
finally, is trustworthy.

Testing for yourself is the *ONLY* way to know for sure.

On Wed, 22 Oct 2014 16:59:42 -0700, Oregonian Haruspex
wrote:

On 2014-10-22 03:58:55 +0000, Eric Stevens said:

On Tue, 21 Oct 2014 16:13:26 -0700, Oregonian Haruspex
wrote:

On 2014-10-17 03:01:34 +0000, Eric Stevens said:

On Thu, 16 Oct 2014 19:13:34 -0700, Oregonian Haruspex
wrote:

On 2014-10-16 00:12:27 +0000, Eric Stevens said:

On Wed, 15 Oct 2014 16:23:11 -0700, Oregonian Haruspex
wrote:

On 2014-10-15 20:54:54 +0000, Eric Stevens said:

So you are saying that currently the best way to enlarge an image with
interpolation is getting someone else to do it. Now, say that 'someone
else' approached you and asked you what is currently the best way to
enlarge an image with interpolation, what would your answer be?

Listen, before the ink shoots out of the nozzles the printer is going
to be doing its own transforms in both the geometric and color domains
anyway, and it's just possible that the fellas who designed its
mechanism and firmware knew what they were doing, no? I mean it's also
possible (but not likely) that things might not turn out that great,
but you can't just take it as a given that in-printer interpolation is
worse than doing it yourself, because doing enlargement yourself in
software is no guarantee that this combined with the printer's own
transforms that it ALWAYS applies as a natural part of the printing
process won't make things worse in the end.

Each printer has its own character and you have to figure out the best
workflow afresh if you change printers.

Hopefully, if you have set things up correctly yourself, the printer
has to do the minimum of adjustment and if your color management is
correct you should end up with what you saw on the screen.

Well it's not like the printer has to do a minimum or maximum of
adjustment, or even that you know what's going on. My only points are
that 1) doing enlargement yourself is not always going to be better and
2) the only way to discover this for any given printer is
experimentation.

There is some very good software out there for rescanning images and I
expect you would have to have a very (very) good printer to do better
than the better available software.

You "would expect" but who knows? Why not try it and find out? After
all, if you are looking to know for sure this is the only way.

I have a *stack* of A2 prints with which I have been experimenting to
answer this question. I gave up leaving it to the printer years ago.
In any case, the question has been answered by others. I don't know of
any book dealing with the production of high quality prints which
recommends just leaving it to the printer.

What sort of methodology have you been using in your experiments? It'd
be interesting to hear about your test setup and what your metrics are
that would lead you to conclude that one print is better than another.

A long series of comparitive prints with their realtive merits
assessed by eye.
--

Regards,

Eric Stevens

On 2014-10-23 08:30:41 +0000, Eric Stevens said:

On Wed, 22 Oct 2014 16:59:42 -0700, Oregonian Haruspex
wrote:

On 2014-10-22 03:58:55 +0000, Eric Stevens said:

On Tue, 21 Oct 2014 16:13:26 -0700, Oregonian Haruspex
wrote:

On 2014-10-17 03:01:34 +0000, Eric Stevens said:

On Thu, 16 Oct 2014 19:13:34 -0700, Oregonian Haruspex
wrote:

On 2014-10-16 00:12:27 +0000, Eric Stevens said:

On Wed, 15 Oct 2014 16:23:11 -0700, Oregonian Haruspex
wrote:

On 2014-10-15 20:54:54 +0000, Eric Stevens said:

So you are saying that currently the best way to enlarge an image with
interpolation is getting someone else to do it. Now, say that 'someone
else' approached you and asked you what is currently the best way to
enlarge an image with interpolation, what would your answer be?

Listen, before the ink shoots out of the nozzles the printer is going
to be doing its own transforms in both the geometric and color domains
anyway, and it's just possible that the fellas who designed its
mechanism and firmware knew what they were doing, no? I mean it's also
possible (but not likely) that things might not turn out that great,
but you can't just take it as a given that in-printer interpolation is
worse than doing it yourself, because doing enlargement yourself in
software is no guarantee that this combined with the printer's own
transforms that it ALWAYS applies as a natural part of the printing
process won't make things worse in the end.

Each printer has its own character and you have to figure out the best
workflow afresh if you change printers.

Hopefully, if you have set things up correctly yourself, the printer
has to do the minimum of adjustment and if your color management is
correct you should end up with what you saw on the screen.

Well it's not like the printer has to do a minimum or maximum of
adjustment, or even that you know what's going on. My only points are
that 1) doing enlargement yourself is not always going to be better and
2) the only way to discover this for any given printer is
experimentation.

There is some very good software out there for rescanning images and I
expect you would have to have a very (very) good printer to do better
than the better available software.

You "would expect" but who knows? Why not try it and find out? After
all, if you are looking to know for sure this is the only way.

I have a *stack* of A2 prints with which I have been experimenting to
answer this question. I gave up leaving it to the printer years ago.
In any case, the question has been answered by others. I don't know of
any book dealing with the production of high quality prints which
recommends just leaving it to the printer.

What sort of methodology have you been using in your experiments? It'd
be interesting to hear about your test setup and what your metrics are
that would lead you to conclude that one print is better than another.

A long series of comparitive prints with their realtive merits
assessed by eye.

I bet you could do better than that, and spend less money on ink too.

On 22/10/2014 00:09, Oregonian Haruspex wrote:
On 2014-10-17 23:00:42 +0000, Eric Stevens said:

If you ask the printer to print twice the size it's almost certainly
going to do it by printing twice as many pixels.

"Almost" certainly? How about "possibly" - after all we really don't
know and it depends on the individual printer, and possibly driver
version as well.

And in some cases how you send the image to the printer. I recall one
very expensive and now elderly dyesub printer that with drag and drop to
printer on original drivers would for some inexplicable reason
downsample the image. The driver was later fixed.

The effect was rather curious in that you got hard edged square pixels
of exactly the right solid colour. It was obvious something was wrong
but it required closer inspection to see exactly what.

What it will have to
do is create a new pixel between each of the original ones. If the
values of the originals were A and B, it is almost certain that the
values assigned to the new one will be (A + B)/2. That might be OK but
if A and B deliniated an edge they would be markedly different and the
new intermediate pixel would soften that edge. That almost certainly
is not what you want.

It might be that the people who do driver development have been
instructed to do just this. It might also be a little bit more
complicated than that - interpolation with edge detection and sharpening
has been a thing in image processing software since the 1980s. The
technique may be applied, or not. Again the only way to know is to test
it for yourself. This isn't like an argument about angels dancing on
the head of a pin, this is something that a person who has an inkjet
printer can actually test and see.

They are usually slightly smarter than Eric's description in that they
usually work across the 4 or 8 nearest neighbours to interpolate pixels
but the odd printer driver may just do pixel replication for upscaling
so test before you commit to a large print. A few cleverer ones might do
edge preserving interpolation but I wouldn't bet on it.

The matter is further complicated by the fact that (except at the
edges) each pixel has 8 neighbours in the first surrounding layer and
16 neighbours in the layer surrounding that. Interpolating the values
of new pixels is both a mathematical nightmare and computationally
heavy. It's better to leave the up-scaling/rescanning of your image to
purpose designed software and only rely on the printer driver to
translate the resulting pixel values to the right pattern of droplets
on the paper.

But how do you know that this all the printer is doing? Without
checking, you don't. It might be that image scaling algorithms inside
the printer have been designed with some attention and care, and that
they can do an excellent job. Who knows? Only the people who check and
see.

Given the state of the user interface on modern "smart" printers I very
much doubt that they have taken that much care on upscaling. You are
always at the mercy of how the printer decides to lay the ink down. Some
patterns can show up there if you have the right sort of image.

--
Regards,
Martin Brown

| I bet you could do better than that, and spend less money on ink too.
|

You seem to be the only person interested in
finding printers that prep images better than
one can do in an editor, so why don't you run
these tests yourself rather than repeatedly
asking other people to do it?