Pinging Alan, Mike and David: more info on the Iwo Jima picture enlargements

Today Mike Warren spoke these views with conviction for
everyone's edification:

All Things Mopar wrote:
Would you or anyone just lurking be at all interested in my
posting the full text of the several E-mails I exchanged with
the fellow who did those 1,600% image increase's of my
Mt.Suribachi picture?

I'm also interested.

OK, everyone, I think we've got enough critical mass here. I'll
start pulling this stuff together. Look for it later this week.

--
ATM, aka Jerry

All Things Mopar wrote:
Today Roger N. Clark (change username to rnclark) spoke these
views with conviction for everyone's edification:


All Things Mopar wrote:

Today Roger N. Clark (change username to rnclark) spoke
these views with conviction for everyone's edification:



Enlarging and shrinking an image requires interpolation.
ALL interpolation algorithms have artifacts, at least all
I've seen. If you compare your original small image with
the largest one, you can see artifacts. I'll point out
one.


Resampling using any algorithm, interpolation or simple
pixel resize, can and does produce damage. But the term
"artifact" first came to prominence during the early
experiences with JPEG, after people first saw the damage
from over compression.

Artifacts in digital imaging is much older than jpeg. I
learned of it in the 1970s in graduate school. Yes digital
imaging was being done back then, pre CCDs. I was using a
vidicon, 256x256 if I remember correctly, system that
digitized with electronics filling 6 feet of rack space,
and a fifty pound camera head, cooled to dry ice
temperatures.


Being that I'm a visual sort of a guy, I don't know how to
use the analytical tools in PSP 9 to examine an image, and
I don't parlez vous PS CS. So, I'm very curious to know
how you can tell that your examples above were caused by
interpolation and not the more likely cause, over
compression at some point, and/or multiple
open/edit/re-save cycles.

Without knowing the exact techniques and algorithms used in
the processing, one can't be sure of where things were
introduced versus where they were exaggerated. In my
experience with scientific interpolation, including writing
imaaging interpolation algorithms, it is my experience that
there is no perfect interpolation algorithm for this type
of problem. What is the likely cause is that the
interpolation in the upsizing is imperfect (causing the
fundamental artifact) which the sharpening and other
processing steps enhanced. Some algorithms invent data
(e.g.fractals) which may look good in many cases, but is
scientifically incorrect, thus all the added "information"
is artifacts. Other algorithms try and do little
"inventing" linear, cubic spline and others, but these
cause other artifacts. One hopes the artifacts have
minimal detriment to the final image appearance.


Not exactly taking you on, but the above seems like spin to
me. A minute ago you were lecturing me on why artifacts are
created simply by interpolation, rather than by
overenthusastic JPEG compression. But you still haven't shown
anything quantitative to refute my definition of the term.
Could you perhaps post some examples that either support or
refute the various opinions expressed herein?

As to "looking good but being scientifically incorrect", this
is /exactly/ my point. Since normal people only look at and
print images, not examine them mathematically except while
post-processing their scans or digitals, the point you're
making is not only moot, but irrelevant.

I got into this thread only to point out that there are
methods, which you say you've personally implemented in
software, that can easily defy the "rules" of either upward or
downward resizing. I still like the bigger images but haven't
been able to track onto the techology used to create noise-
free versions, but I still sleep fine at night with only a
postage stamp picture of my father.


By scientifically incorrect, I mean "not real." If it is not
real, the deviation from reality can manifest itself in multiple
ways, from not visible, but scientifically inaccurate to useless
to very visible and still scientifically useless.

You can do a google search for: digital image resampling artifacts
You will get MANY hits.

Here is a PhD thesis on the subject from 2002:
http://bigwww.epfl.ch/vandeville/phd/
The 1st paragraph of the abstract says:
"Digital image-processing systems are becoming more and
more important nowadays, pushing away analogue techniques.
For example, the advent of advanced digital cameras and
the complete digital workflow in prepress industry.
Digital image processing is inextricably bound up with
sampling on lattices. Consequently, the need for techniques
to convert images and video from one lattice to another
one is as old as digital image processing itself.
Nevertheless, resampling is still often treated
unsatisfactory. The usage of simple but widespread
techniques such as nearest neighbour interpolation and
bilinear interpolation, gives the wrong impression that
resampling is an easy process. The frequent appearance
of artifacts in images and video as a direct consequence
of resampling, proof the importance of paying sufficient
attention to these techniques. For example, most PostScript
RIPs use bilinear interpolation and introduce undesirable
artifacts such as moire patterns. Video formats for television
still cling to the interlaced scanning format resulting in
visual artifacts and unsuitability for progressive media
such as flat panel displays. An appropriate resampling step
is needed here too."

A science paper by the above author:
http://bigwww.epfl.ch/vandeville/pub.../phdsymp01.pdf

Image Restoration Using the Damped Richardson-Lucy Method
http://www.stsci.edu/stsci/meetings/...er_damped.html

Google: Richardson Lucy deconvolution artifacts
You will see many hits.

While I do not discuss artifacts on this page, the page does show
an increase in detail the R-L method achieves over standard
photoshop methods:
Image Restoration Using Adaptive Richardson-Lucy Iteration
http://clarkvision.com/imagedetail/image-restoration1


Back to your photo. The original artifact I pointed out,
the smile-cheek merger is most likely caused by the resampling
algorithm(s), and then was enhanced by the sharpening
methods.

Roger

All Things Mopar wrote:
[]
OK, everyone, I think we've got enough critical mass here. I'll
start pulling this stuff together. Look for it later this week.

Many thanks, Jerry.

David

Today Roger N. Clark (change username to rnclark) spoke these
views with conviction for everyone's edification:

By scientifically incorrect, I mean "not real." If it is
not real, the deviation from reality can manifest itself in
multiple ways, from not visible, but scientifically
inaccurate to useless to very visible and still
scientifically useless.

I get 1,270,000 hits for "cognitive dissonance", mebbe you
should look to see what that means, because you've got a fatal
dose of it.

Over and out, your 5 minutes are up.

You can do a google search for: digital image resampling
artifacts You will get MANY hits.

Here is a PhD thesis on the subject from 2002:
http://bigwww.epfl.ch/vandeville/phd/
The 1st paragraph of the abstract says:
"Digital image-processing systems are becoming more and
more important nowadays, pushing away analogue techniques.
For example, the advent of advanced digital cameras and
the complete digital workflow in prepress industry.
Digital image processing is inextricably bound up with
sampling on lattices. Consequently, the need for techniques
to convert images and video from one lattice to another
one is as old as digital image processing itself.
Nevertheless, resampling is still often treated
unsatisfactory. The usage of simple but widespread
techniques such as nearest neighbour interpolation and
bilinear interpolation, gives the wrong impression that
resampling is an easy process. The frequent appearance
of artifacts in images and video as a direct consequence
of resampling, proof the importance of paying sufficient
attention to these techniques. For example, most PostScript
RIPs use bilinear interpolation and introduce undesirable
artifacts such as moire patterns. Video formats for
television still cling to the interlaced scanning format
resulting in visual artifacts and unsuitability for
progressive media such as flat panel displays. An
appropriate resampling step is needed here too."

A science paper by the above author:
http://bigwww.epfl.ch/vandeville/pub.../phdsymp01.pdf

Image Restoration Using the Damped Richardson-Lucy Method
http://www.stsci.edu/stsci/meetings/...dings/whiter_d
amped.dir/whiter_damped.html

Google: Richardson Lucy deconvolution artifacts
You will see many hits.

While I do not discuss artifacts on this page, the page
does show an increase in detail the R-L method achieves
over standard photoshop methods:
Image Restoration Using Adaptive Richardson-Lucy Iteration
http://clarkvision.com/imagedetail/image-restoration1


Back to your photo. The original artifact I pointed out,
the smile-cheek merger is most likely caused by the
resampling algorithm(s), and then was enhanced by the
sharpening methods.

Roger

--
ATM, aka Jerry

Today David J Taylor spoke these views with conviction for
everyone's edification:

All Things Mopar wrote:
[]
OK, everyone, I think we've got enough critical mass here.
I'll start pulling this stuff together. Look for it later
this week.

Many thanks, Jerry.

David, you and I occasionally clash, but we're usually in
agreement on at least on the majority of the paragraphs of the
same pages. grin

I jumped in here on a lark, and got hooked. I gave up a year ago
trying to understand this stuff, but now my curiosity is piqued
again, so, altruistically, I have a vested interest in
understanding the theory and practice and maybe an opportunity
to get free or low-cost solutions to problems that've long
plaqued his normal PSP and PS CS folks.

"Inquiring minds want to know"...And now, a quote from one of my
better boss's - "nothin' to it but to do it!".

--
ATM, aka Jerry

All Things Mopar wrote:
I get 1,270,000 hits for "cognitive dissonance", mebbe you
should look to see what that means, because you've got a fatal
dose of it.

Over and out, your 5 minutes are up.

What an absolute jerk. You asked for help and I tried.
If you want to believe that resampling does not produce
artifacts, fine. You are now in my killfile.

Roger