[WikiEN-l] Wikimedian image restorations exploited on eBay

[Durova]

Restoration is inherently interpretive.  Consider something simple: a
newspaper cartoon in black and white.  There are many possible whites; which
do you select?  Do you retain or eliminate paper grain?  Older illustrations
are often imperfect by a few tenths of a degree, so when the border isn't
quite rectangular what rotation do you choose?  Do you crop wider to
compensate or do you crop out the border itself?  When you detect an obvious
printing error such as an uninked spot within a straight line, do you fill
it in or do you retain the empty spot?  If you retain that spot when it
looks like a printing error, what do you do when ink rubs away from the page
after printing?  Or when you're not sure of the cause?

The two most prolific Wikimedians in this area are Shoemaker's Holiday and
myself, and although we often work together we also have longstanding
philosophical differences that reflect in our featured picture galleries.
The most obvious of these regards color balance.  A more interesting debate
concerns nineteenth century etchings and engravings (it's interesting to
us--might bore the rest of you to tears).

People who rely on tools and plugins don't avoid interpretion; that only
delegates the interpretive work to a computer program.  There's an example
from my bookshelf which, fortunately, also happens to be available via
Google Books preview.  Scroll to the Texas saloon on page 11.

http://books.google.com/books?id=SNoNlmvJQy4C&printsec=frontcover&dq=digital+restoration&ei=-_mzSqbdNqKIkATfoamJBA#v=onepage&q=&f=false

This author is very helpful in some other respects, but his reliance on
plugins is a liability here.  The software has made choices with the
building facade which are clearly wrong: real windows don't morph into
puddles.  Enough of the right window remains visible to show that it is a
duplicate of the left window.  A better reconstruction would borrow data
from the intact window.  The vertical lines of the facade planks can be
rebuilt in a similar way.  Shadows on the facade and men's clothing gives a
trustworthy measure of the sunlight's angle, direction, and intensity.  That
would serve as a reference for distinguishing and correcting brightness
variances that result from stains.

Of course if this were a Commons upload the edits would be documented in
detail on the image hosting page, the unrestored file would be uploaded
under a separate filename, and both file descriptions would link to each
other for cross reference.

-Durova

On Sat, Sep 19, 2009 at 5:28 PM, Phil Nash <pn007a2145 at blueyonder.co.uk>wrote:

> I agree from this, and your previous post, that restoring historical images
> can be a difficult process, particularly when the images themselves may
> have
> originally been pure factual journalism rather than having a polemical
> purpose, although in my experience, that is more allied to the commentary
> attached than the image itself. In the case you cite, processing an image
> may well involve some interpretation of the depiction, and you rightly
> point
> out some of the pitfalls involved. Absent the intention of the
> photographer,
> who may not even have considered how his image may have been used (as long
> as he was paid), making assumptions I believe to be unhelpful, and even
> Original Research. All this convinces me that image restoration should be
> limited to correcting obvious physical defects in the source, and not going
> beyond that. I am not in any way criticising those who do this (after all,
> I've done it with my own images, although I do know what I intended when I
> created the image), bur I do believe that restoration should not blur into
> interpretation.</ramble>
>
> --
http://durova.blogspot.com/