Happy Halloween

As a Wiki Witch I can't really let the day pass without a note about today's picture of the day.  This one was created in 1892 to mark the two hundredth anniversary of the Salem witch trials.  The revisionism here is really delightful: an innocent defendant, clad in white, calls down divine lightning to smite her accusers.  Manacles spring open in the background while the prosecutor drops to the floor and the judges rise to their feet in amazement.

Possibly next year's Halloween featured picture will be one of the Édouard Manet illustrations for "The Raven."

Happy Halloween.

A matter of perspective

Today's challenge began when GeraldK asked for help getting a featured picture about the Winter Olympics.  He's active with the Olympic Games project and is hoping to have something run on Wikipedia's main page in time for the 2010 Olympic Games.  The Bundesarchiv material he suggested was far too small for consideration: ranging from 21 KB to 51 KB.  Normally 10 MB would be the minimum for my work.

So Staxringold found a 32 MB poster from the Works Progress Administration about the bobsled run at Lake Placid, New York.  Lake Placid hosted the Winter Olympics in 1932 and 1938; its bobsled run was converted to public use after the 1932 Olympics.

So we've got a technically adequate file that fits the topic.  It comes with a problem: the dimensions aren't rectangular like they ought to be.  This is how it looks after 0.7 degree counterclockwise rotation.

This demonstration crop leaves a little background to help illustrate the problem.  If you look carefully enough, the caption itself is slanted in conjunction with the distortion.  There's an elegant solution to this in the form of a software option which mainly serves the different purpose of perspective control.

 When perspective distortion is a problem it usually appears in architectural photography.  Back in the days of film photography people corrected for it with a view camera or a perspective control lens.  In the digital era this usually gets corrected with software.  The same software controls can help the Lake Placid poster too.

In Photoshop this is called perspective cropping.  Perspective cropping allows the editor to drag corners into non-rectangular shapes.  Then the software performs calculations to generate a rectangular result.  To access the perspective crop, select the regular cropping tool and drag a crop selection.  A toolbar option will appear above with a check box for "perspective".  Check that box and the perspective adjustment becomes possible.  In our situation that means aligning cropping tool with the corners of the poster.  Just place one corner of crop selection directly onto each corner of poster, then crop. GIMP software users can get a similar effect from the Perspective Tool.  Below is the result after additional edits.

Much more satisfactory and probably fairly close to how the poster looked when it was new.  If the distortion had been more severe we might have supplemented the perspective crop by separately rotating the poster caption.  But this was a fairly mild instance and the perspective crop did the job.

What causes this distortion, you may wonder?  Many of the items from the Library of Congress poster collection were photographed for preservation purposes during the film era.  These posters were already old and fragile at the time of the photography and they may not have laid perfectly flat.  A slight rise or buckling resulted in the distortion.  So if the digital file was made from the slide film rather than from the original, the same problems get translated into digital format.

----

Photo credits:

http://commons.wikimedia.org/wiki/File:Cathedral_Notre-Dame_de_Reims,_France.jpg

http://commons.wikimedia.org/wiki/File:Cathedral_Notre-Dame_de_Reims,_France-PerCorr.jpg

Holidays and special occasions

"You're doing Valentine's Day in October?"  A friend commented in amazement when I shared the half-finished work.  Well yes, actually it's best to work on these things months in advance.  Valentine's Day 2008 taught us that.  From the Wikipedia Signpost report of March 13, 2008:

After being featured and de-featured in February, Image:Love or dutyb.jpg, an 1871 chromolithograph by Gabriele Castagnola, has been re-featured. A suggestion by Spikebrennan for a Valentine's Day featured picture led Durova to locate the image at the Library of Congress archives, restoring it as Image:Love or dutya.jpg, where it passed featured picture candidacy on a unanimous vote; however, some users complained that the vote violated usual procedure. In order to feature the picture on the Main Page on Valentine's Day, the vote ran for just over two days, while most nominations run for about seven days. After being featured on the main page, these procedural issues led to a nomination for delisting, which resulted in the image's de-featuring. Afterward, the image was renominated as Image:Love or dutyb.jpg, where it passed on a second unanimous vote. This is believed to be the first time that a featured picture has been nominated, promoted, highlighted on the main page, delisted, and repromoted in less than 30 days.

That's quite a journey for an innocent little depiction of a nun and a painter...well, almost innocent...

The featured picture regulars reacted in good spirit.  It was a hoot, but it was also a learning experience.  Holidays and special occasions really do need advance preparation.

Other editors are welcome to contribute holiday and special occasion featured picture candidates.  Chinese New Year, Mardi Gras, and April Fool's Day 2010 are wide open.

Good news about Booker T. Washington

Early this year I blogged about the unsuccessful search for a quality portrait of Booker T. Washington.  It's a good thing to retry searches occasionally because the Library of Congress keeps adding to its digital selections.  Now they provide a wonderful image of him.  It has good composition, has been well curated, and is technically better than Wikimedia's software can accommodate.  This cropped version is 126MB in 16 bit TIFF format.  Wikimedia uploads are limited to 100MB and 8 bit TIFF files.  We'll have to reduce the technicals a little bit to get this uploaded.  So the media editors will need to start tugging on developer sleeves.

This is a good problem.

At full resolution quite a bit of work remains to be done.  Glass plate photography is a very delicate medium and almost always needs attention.  The technical condition of this one is better than average.  Here's a closeup from part of the chair.

Cheers to Xavexgoem; he's working on this now.  I had gotten a start on the Washington restoration and then been pulled away by other projects.  Had been feeling guilty about not returning to this sooner.  Xav was looking for something to do last night and loves the portrait as much as I do.  We decided to trade off and share the work.

This is what Wikipedian collaboration is all about.  Now does somebody have a talent for tugging at developers' sleeves?  It would be a shame if the full version of this restoration existed on only two people's computers.

Mountains and molehills

There's nothing like a rare photograph of an important historic event.  The one above is the eruption of Mount Vesuvius in 1872.  The shot is magnificent and the scan is superb: 217MB at full resolution.  After rotation and cropping the uncompressed version comes to 93MB--just within the upload limit for Wikimedia Commons.

As you can probably tell, though, this image needs a lot of work.  At full resolution it's really daunting: thousands of spots plus substantial areas of chemical damage at the slopes and coastline.  So today's post is about what to do with images that have a lot of problems.

First, this isn't a beginner project.  Be realistic about your time and skills.  This is the sort of restoration one works up to doing, and even after hundreds of others I feel like I'm on a fool's errand for trying it.  A closer look may convey that more clearly.

This is one small bit of coastline.  The full version as it sits on my computer now is 6684 x 4857 pixels; I've been working on it with a tool 9 pixels in diameter.  That means a whole lot of labor and some of the choices will be difficult, such a what to do about the patch of chemical damage at lower right.  Most novices approach this sort of problem by prioritizing the large areas; I do the opposite and work on the small problems first.  See below.

Eventually the large problems such as the chemical stain will have to be patched in with repairs from other parts of the image.  So it makes perfect sense to start by resolving the less serious problems at those other areas.  I'm lazy; I'd rather resolve a problem once than copy it somewhere and fix it twice.

Also, clearing up the little stuff gives time to think about the hard part.  This particular stain covers coastline, part of a town, part of the Vesuvius slope, and also--at its very bottom--part of a sailboat.  That's going to be a monster to resolve.

So the first question to ask before even deciding whether to do a restoration is to look for the hardest problems and evaluate whether it's within one's skill set to address them.  I think this is resolvable, although not necessarily sure exactly how.  One of the advantages of clearing out the minor problems first is that it eliminates distractions and makes it easier to focus on the difficult issue.  Another advantage of this approach is that it gives the subconscious time to work out a resolution.

My subconscious is much brighter than I am.  It makes me look smart.  I do the simple routine fixes, the subconscious goes to work on the hard problems.  Then once the subconscious has pondered the matter for days I steal its solution and take credit.  It's a nice arrangement for me; the poor exploited subconscious has nowhere to complain.

On a more serious note, here's a closer view of the unaltered and partially restored slope.  You can tell that I haven't resolved everything yet--just working conservatively from the smaller and simpler problems up to the larger and more complex ones.  I usually take a particular tool and adjust it to particular settings, then sweep through the image in a grid pattern doing everything that tool setting is appropriate for.  After several sweeps and several different tools and settings the image will approach completion, and be ready for specialized attention to the larger problems.  The tool used here was the healing brush at 9 pixel radius, 100% hardness.

One last thought before parting: Vesuvius is a gorgeous location where humans really are better off visiting than living.  You've heard of Pompeii?  That was Vesuvius's doing.  Maybe a great find for archaeologists, but I'd rather contribute to posterity by digitally restoring images than by curling into a fetal position, succumbing to poisonous fumes, and getting buried in volcanic ash.

Vesuvius wakes up and gets temperamental on a regular basis, but not quite frequently enough for the human lifespan.  Its last significant eruption was 1944.  Have a look at its slopes today: dunno about you, but I sure 'n heck wouldn't want to live there.

Maybe the utility of restoration work is to serve as a reminder.

Chromolithography

For digital restoration work it is important to develop a working understanding of historic media.  A given medium behaves in predictable ways even when applied in very different styles.  Two restorations I worked on yesterday are examples of that.  The medium is chromolithography.  The example above is a World War II propaganda poster, which an editor who goes by Staxringold selected in part because of its racial overtones.  Below is an example nearly a century older: an 1847 print about Irish political leader Daniel O'Connell.

The Library of Congress doesn't specify which medium O'Connell's portrait is.  I can't specify it onsite because that would violate Wikipedia's policy regarding original research, but to my eye it's very clear that this is either a hand painted lithograph or a chromolithograph.  One gets to know these media after a while; one learns how they behave.  Limestone produces a recognizable grain upon an image.  Despite the vast differences in content and style, both of these files bear technical similarities.

There's a lot to be written about how chromolithographs behave during digital restoration.  One place to start is the slightly uneven color application which is inherent to this process.  Because of that, the two principal tools of digital restoration--the clone stamp and the healing brush--have to be used judiciously.  One problem that's important to avoid is the tile effect.  With the example below the same two-part motif appears almost in a diamond pattern.

The problem here is caused by repetitive use of the same source point upon multiple destination points.  Fortunately it can be corrected in a large open space by resampling from different source points and reapplication over the affected area.  The ideal to aim for is a natural randomized pattern like that of real limestone grain.

Renomination

One of the hard things about the consensus process at the wiki is that sometimes one comes away thinking "this one went the wrong way".  This particular restoration went up last January and got unanimous support, but only three people reviewed it which wasn't enough for promotion.

Two days' hard work went down that way, and although the subject, Edward M. Thomas, was a war hero there wasn't quite enough information to start a separate biography for him; he was one of the Tuskegee Airmen.  So the image remained at photographer Toni Frissell's biography and a couple of other pages, and I moved on.  Which wasn't easy after having read the text of Captain Thomas's Distinguished Flying Cross award; he's the sort of person who deserves to be remembered.

Which was why it felt so good yesterday when a Wikipedian who goes by the username NuclearWarfare found it and decided to give it a renomination.  Very pleasant surprise to see him take an interest in it, and even more so to see the editor reviews.  Thank you very much!

Susan Sarandon doesn't know what Wikipedia is

Andrew Sullivan of The Atlantic Monthly has picked up on David Shankbone's recent blog post about an encounter with Susan Sarandon, where Sarandon said she didn't know what Wikipedia is.

Well good for her!  It's refreshing to find a celebrity who is not fussed about her Wikipedia biography and not vain about the portrait that runs at the lead.  David took the shot at right last year at the Tribeca Film Festival. 

She's one of the best actors in the business, she's been making films since 1970, and she looks fantastic in more ways than one.

Kudos, Susan.  Keep it real.

Anniversary

Today marks my four year anniversary as a Wikipedian.  Another featured picture promotion came in last night, so I have 333 featured credits now.  Perhaps in another four years I'll have 666 featured credits and sell Wikipedia to the devil... ;)

Raising the color bar

During the twentieth century one technique for archival preservation was photography on high quality film  This could be invaluable in case the original document later gets lost or destroyed.  Eastman Kodak, which made the film, provided a grayscale and color bars.  And that is the source of today's dilemma.

It's a reasonable first reaction to the presence of color bars and grayscales to apply them mechanically and call the result original.  But looks can be deceiving: in digital restoration it's always important to think about which sort of data is useful and which isn't.

The key thing to bear in mind is that the color bars represent data at the time the photograph was taken.  They do not yield data on how this poster looked when it was new.

This World War I poster was created in 1916.  The earliest color Kodak film was produced in the 1930s.  Color film rephotography remained in use until digital technology replaced it.  So any World War I poster that got preserved by this technique would have undergone at least two decades of decomposition before the photograph was made.

An unknown amount of distortion occurred prior to photo reproduction: if you've ever spent a summer on the Eastern Seaboard, consider the damage that twenty humid summers could inflict.  Especially without air conditioning.

But twenty years is the bare minimum here: the date of this particular reproduction was not recorded.  The bars and grayscale might be somewhat useful if we knew what year this was taken: 1946?  1966?  1986?  That time frame is too large to be useful, which is why I crop out these bars and ignore them.  That may sound radical, but yes I ignore them.

The reason becomes obvious when working with this material: substantial decomposition had already taken place.  The lower half of the poster is more deficient in blue than the top half, with additional localized problems such as the pronounced vertical yellow band at far left.  Chromolithography from this era can behave that way as it ages.  The edges had dried and darkened more than the center, which required radial gradient masking to correct, and the brittle poster did not lie flat for reproduction: it needed a perspective crop, then half of the caption had to be rotated separately in order to recreate the original alignment, and then separate corrections had to be applied for the effects of uneven lighting.

So it wasn't until today when another editor at Wikipedia's featured picture candidates insisted that the gray scale bar gives definitive data on how this image looked when it was new, that I located a more specific demonstration of why it does not.  First, here's the gray scale bar.

 

Do you notice something odd about that gray scale and color bar?  Let's look in closer.

Copyright Eastman Kodak Company, 1977.  That's right: this poster was at minimum 61 years old when the photographic copy was made.