Thursday, September 24, 2009

A sidelong glance

Wikipedia's internal manual of style says portraits shouldn't look away from the text.  At the actual biography this one looks away.  A lot of them do.  In this instance that choice especially apt; this is George Atzerodt.  The night Abraham Lincoln was assassinated, Atzerodt had plans to assassinate Vice President Johnson.  Atzerodt was an associate of John Wilkes Booth.  Except after Atzerodt arrived at the hotel where Johnson was staying he didn't go upstairs; he stopped at the bar.

And got drunk instead.

Atzerodt could have notified the authorities and perhaps saved Lincoln's life; he didn't.  Atzerodt could have backed out of the conspiracy; he didn't do that either.  He could have tried to kill Johnson.  Instead he went on a bender.  This portrait was taken after his arrest.  The irons on his wrists are slightly out of frame.   He went to the gallows after having earned the respect of nobody.

The look on his face seems to summarize the man: intensely avoiding eye contact.

The Atzerodt portrait was a restoration I did several months ago.  There's another set of portraits from the same group that's more challenging.  Alexander Gardner's portraits of Lewis Powell have an informality that was rare for the era.  Powell was twenty-one years old.  He has the build of a college athlete, except the manacles are visible and he leans against the gun turret of a Monitor class warship.  Powell had stabbed Secretary of State William Seward repeatedly in an attempted assassination.  It was a lucky chance that Seward survived.  When Powell assaulted Seward, Seward was bedridden and wore a neck brace.  The brace prevented Powell's knife from reaching the jugular vein.

Two years later Seward organized the purchase of Alaska.

A neck brace and some drinks at a hotel bar changed history.

Powell is hard to crop.  It's also difficult to select just one portrait; he's very natural in front of a camera.  So I haven't worked on him yet.  Maybe someone else will beat me to it.  There are a lot of worthwhile images out there.

Friday, September 11, 2009

Let's have a TIFF

Wonderful news from the Wikimedia Technical Blog today: Wikimedia Germany is developing full TIFF support for Wikimedia Commons with implementation scheduled for November.

Earlier this year the WMF developers enabled TIFF uploads to Commons for the first time, mostly at the request of the growing restoration community.  TIFF is a file format that's widely used for many real world applications including archival digitization, but most web browsers do not display it and WMF software is unable to render it.  The new development will make WMF more TIFF-friendly.

TIFF support facilitates file uploads directly from museums, libraries, and other cultural institutions.  It also aids wiki-style collaboration among image editors.  The most popular format for online display (JPEG) is a compressed format that undergoes progressive deterioration from repeated saves and edits.  TIFF support helps media editors improve on each others' work without compromising quality.

Although TIFF is not the only uncompressed file format hosted  at Wikimedia Commons (PNG is another), TIFF's dominance in off-wiki uses means that full TIFF support adds long term scalability to our acquisition of encyclopedic media files.

Kudos to WMF Deutschland!

Thursday, September 10, 2009


Wikipedian  MZMcBride recently launched a new tool that reports how many people watchlist a given page.  This has interesting uses.  Word first reached this direction last night.  Today Jehochman's page update served as a reminder (because, of course, I watchlist it).  Wikipedians being what they are, I soon ran a search on Jehochman whose userspace is being watched by 263 people.  Comparison to Jimbo Wales's user talk (1893 watchers) shows that Jehochman's user space receives 13.9% as much attention.

So to coin a unit of measurement, the centijimbo is 1/100th the number of watchers that Jimbo Wales has.  One centijimbo currently equals approximately 19 watchers.  It's a relative measurement so this could alter as people add and drop watchlist pages.

Here's how several prominent Wikipedians measure up:
The two most lurked Wikipedians other than Jimbo himself appear to be JzG and Raul654, who have 525 and 524 watchers respectively.

One might hold interesting debates over what these numbers mean, which factors affect them, or why the Wikipedian with the highest centijimbo rating has a deleted userpage.
Juliancolton (13.8 cj) has noted that SlimVirgin measures 32.9 centijimbos.

Tuesday, September 08, 2009

A touch of class

One of the most fascinating photographic exhibits I've attended was a show of George Hurrell's portraits at the University of Southern California library.  Hurrell was a master of Old Hollywood glamor.  It's hard to find material of his under free license so here's a link to his iconic portrait of Jane Russell.  Hurrell was a very talented photographer whose legacy is augmented by two very fortunate circumstances: his subjects were among the most beautiful human beings on the planet and Hurrell employed a stellar team of retouchers.  During an era when plate photography was getting supplanted by film for most purposes, Hurrell shot on oversized glass plates; his retouchers would have half an hour to retouch the original negative before it dried.  The results require viewing at close range to fully appreciate: they don't feel artificial. Errol Flynn looks genuine down to the pores on his cheeks, except it isn't a close shave. Those pores look like he never grew stubble at all.

Hurell's work came to mind tonight as I was dabbling with a very different undertaking.  Although it wouldn't be right to beautify a portrait subject for encyclopedic restoration, reconstructing data on photographic prints does bear similarities.  Bad image editing takes on an airbrushed appearance similar to a weak attempt at physical retouching.  Good restoration work on high resolution digital files includes reconstructing photographic grain at damaged areas.  The finished image needs to appear credible.  For Hurrell's studio it wasn't enough to take out stubble; they had to rebuild Flynn's pores.  It isn't enough to take out dirt and scratches; I rebuild grain.

Images that are at the margins of restorability fascinate me.  This one of Cy Young from 1905, for instance.  It's shot through with problems: faded, stained, smeared, dirty, and creased across the middle.  At full resolution it also has fingerprint marks (remember from Honus Wagner how difficult fingerprints can be?) and the Cy Young print even carries a mirror image imprint from a ball point pen.  The print was probably stored with a sheet of handwritten paper facing it, and over years the ink transferred.  It'd be a very tall order to repair the damage.  If an editor gets that far, there's also the dilemma of whether to crop away part of the stadium to get rid of two distracting figures at far right.  Plus it's contrasty and Young's posture is awkward.  But then it's also a wonderful little scene of a professional ballpark in use at the start of the twentieth century.  And it is Cy Young.

This type of image is digital quicksand: alluring enough to want to try, damaged enough to be monumentally hard, and no matter how good the restoration is there's a serious chance that reviewers will reject it at featured picture candidacy due to the underlying compositional flaws.  I specifically asked Staxringold to stay away from it.  And then, being a sucker, I dabbled at it myself.

Here's a detail of one of the more heavily damaged areas.  The wispy horizontal stripes with a bluish hue are surface glare from the reproduction.  The print buckled when it got creased, so digitization picked up undesirable details of the print texture.  Accompanying these lines are a set of fainter horizontal dark bands that received too little light during reproduction.  A fairly large stain is also visible at left.  The challenge is figuring out what to do with this.  It's editable: most of the important data hasn't been harmed directly.  I've resisted the temptation to drag a clone stamp across it because it's important to retain the impression of a crowd scene.

This is where it's come after more effort than I'd care to admit.  Obviously nowhere near completed.  It's been more like an exercise in reconstructing grain and texture: rebuilding the wooden beam, melding the brown stain, reducing highlights on the folds.  Most of this work was done six pixels wide with a healing brush at 200% resolution.  It builds skills to toy around with heavily damaged material.  Like Hurrell's retouchers, the art is to reconstruct something believable in place of the problems.  It's time to set this one aside and think of alternate solutions.  Possibly next I'll separate out the blue channel.  The subconscious can be a wonderful assistant; the subconscious can chew on this task for a while.

Sunday, September 06, 2009

Blue sky speculation

Occasionally a restoration project goes onto the back burner for a while.  Most of this image was routine.  It's a photographic print from the end of the American Civil War that shows the ruins of a railway roundhouse after Atlanta was burned.  The serious problems were in the sky.

The project started as a scan from a high resolution album. A fascinating historic document: at the time when this war happened photographic technology was just reaching a level where serious visual documentation was becoming possible. Quite a few of the prints from this album may become featured pictures eventually. Roundhouses were sturdy structures.  To see this one almost completely demolished is a powerful reminder of the destructive force of that war.

Relatively speaking, skies are easier to fix than other elements on a photograph.  The data beneath the horizon was basically sound so I selected this print as a project.  But this isn't a simple blue sky; those cumulus cloud formations turned out to be quite frustrating.

This sky has several problems: brightness differences, creases, and stains.  It looks like a combination of chemical and physical damage affected the upper center.  A lot of work went into correcting these features and I wasn't confident about the result.

It's a matter of how high to set one's standards.  On every subject there will be a few people who spot inconsistency.  When a television program switches props between scenes, ninety-nine people out of a hundred may notice nothing, but there will also be one person who says, "Wait!  He was wearing a three-quarter helmet when he parked the motorcycle.  But now he's walking into the house carrying a full face helmet.  That makes no sense."

A pure blue sky is easy to repair, but cloud formations are meteorological phenomena.  Would this edit insult the intelligence of a person who understands meteorology?
In true wiki fashion I asked another amateur editor.  Juliancolton isn't a meteorologist, but he has written about twenty featured articles about hurricanes.  That shows either very good judgment or very bad judgment on my part; take your pick.

When Julian received both images he was slow to reply.  At least he wasn't indignant about the edit.  After discussion he agreed it looked credible.

Which is a relief.  But since Wikimedia Commons is a wiki I'll also be uploading a partial edit (without hisotgram or color adjustment) in case someone disagrees.

Saturday, September 05, 2009

Lifting fingerprints

Fingerprints pose challenges to restoration.  Above is a photograph of Honus Wagner, circa 1911, which is available in a 55MB digitized file from the Library of Congress.  It is a pleasure to see institutions provide such high quality copies, yet with material of this age the odds are pretty good that someone has been sloppy about handling the document.  In this instance  the biggest problem is a fingerprint beside the right elbow.

Getting in a bit closer reveals what a difficult issue this is: the fingerprint whorls are roughly perpendicular to the seams and creases of Wagner's uniform and they continue past his uniform to the grass behind him.

What we want to do is retain the meaningful data while getting rid of the fingerprint.  A lazy approach would be to remove the brightest fingerprint highlights.  That won't be good enough: the human brain is excellent at pattern recognition.  Subtle sets of parallel lines will still be apparent.  So it's going to be necessary to recreate the pinstripes on the sleeve, the creases on the pants leg, and the relevant shadows.

Whenever a restoration task seems daunting a good approach is to do the easiest part first.  A little time and familiarity makes it simpler to work out the rest.  So we'll start at far right.  I used the Photoshop healing brush tool at 100% hardness and 9 pixels.  An important trick to getting good results is to change the source sampling point frequently.

Moving left to the edge of the leg, the healing brush becomes more tricky.  The tool's algorithm draws data from outside the selection area, which tends to smudge borders.  Most people switch to the clone stamp in this type of situation.  I recommend around 35% hardness for clone stamping in most restorations, including this one.  The reduced hardness usually avoids the cookie-cutter appearance of novice image editing.

Editors who are more ambitious could try what I did here: it is possible to use the healing brush effectively at borders if the source area and destination area are close matches.  I reset the tool diameter to 5 pixels and sampled from a border section lower on the leg, then applied the tool to the fingerprint marks at the edge of the bright area.  The advantage of this approach is that when it works it yields a very good blend.  If the results aren't desirable at the first attempt, just back out through the history and start again.

Although the clone stamp tool is good for restoration at low hardness, I nearly always keep the healing brush set to 100% hardness.  Anything less gives muddy results.  Here's how the area appears after a few minutes' work.

Now it's time to handle the toughest sections.  Most of the work on this photograph has been done at 200% resolution.  For this part we'll go in at 300% resolution and use both the clone stamp and the healing brush.  The aim is to infer meaningful data about the arm, elbow, shadows, and uniform from the areas that haven't been marred by the fingerprint.  The results won't be perfectly sharp because the photo is no longer sharp at such high resolution.  Part of the goal is to generate a slightly mottled appearance that looks compatible with undamaged areas elsewhere in the photograph.

There's the result at high resolution. Now we'll pull out and see how this blends with the surrounding area.

This is not a complete restoration, just a partial screenshot of the image as it exists on my computer as of this writing.  The fingerprint itself was rather large and this demonstration didn't erase all of it.  Notice the subtle grooves above and below the area that was worked on?  That's what I meant about the human brain being an excellent at pattern recognition.  Other parts of the fingerprint still remain.  This work goes more slowly than ordinary scratch and dirt removal.

Fingerprints are chemically interesting.  From a preservation standpoint that means there's a strong likelihood that oils and dirt will cause chemical reactions on a photographic print.  Those changes can't be brushed or washed away.  Attempts at cleaning often cause additional damage.  When you need to scan a damaged print, might as well leave bad enough alone on the physical item and correct the problem digitally.

So if you have a family member with bad habits at handling old prints, sit them down and show them this post.  Better yet, fire up an image editing program and task them with cleaning up their own mess.  Whether the goal is to preserve baseball history or family memories, it's much less effort to wash one's hands and hold photographic prints by the edges.

Friday, September 04, 2009

Which way is up?

It only takes a moment to sense something off kilter with this shot.  Notice how all the buildings slant to the right?  This image needs counterclockwise rotation.  Many rotations are easy; this one isn't.  The photograph has mild perspective distortion.  No matter how one rotates it doesn't feel correct because the verticals on these buildings aren't quite parallel.  Also the sidewalk slopes upward from left to right.  Decreasing the slant on the structures means increasing the slant on the sidewalk.  So unless an editor is very careful about which cues to follow, something is going to come out wrong.

It helps to know that this is the Brooklyn Daily Eagle storefront, photographed in 1916.  Most of Brooklyn development occurred between the American Civil War and World War I.  This was relatively new construction when it was photographed so we can discount the possibility that significant shifting or settling would have altered the buildings themselves.  Let's look for trustworthy points of reference within the image.

The man descending the stoop uses canes, so he might not stand upright.  But what we can depend on is that the steps would be level.  Looking closely at the base of this stoop confirms my suspicions about the sidewalk: notice how the base at the left edge of the stoop is nearly twice as long on its left side as on the right side. This was compensation for sloping terrain.

Here's how that stoop appears after rotation and spot removal.  These weren't the only lines of reference when I rotated; I also went by the building verticals and roughly split the difference in terms of angle distortion.

Here's a glance of the whole image after rotation, cropping, and several lengthy rounds of dirt and scratch removal.

It's better than things started, but the distortion still leaves an uncomfortable sense that the lamppost is going to keel over onto something and the French restaurant at right looks as if it is leaning into the Eagle offices.  This keystone effect is subtle, but it's there.  The solution is to perform a mild perspective crop.

Here's the end result.  Added a touch of curves adjustment to bring out contrast in the midtones, plus localized brightness and contrast corrections.  The automobile at far right no longer fades toward the edge of the frame.  Let's hope those ancient cars had good brakes because the driver didn't angle the wheels toward the curb.  Otherwise, nothing looks askew.

Thursday, September 03, 2009


Surprising news arrived today. The discovery of human remains from the Wounded Knee Massacre aftermath restoration has been incorporated into the official program notes of an exhibit at the Montreal Museum of Fine Arts, item number 76. The Library of Congress also credits me in its updated bibliographic record. The museum is exhibiting an unedited version of the image; possibly the presence of the bodies was a factor in its selection.

The individual who shared the news does not appear to have a connection to the museum. Would love to make contact with Quebecois Wikimedians who could follow up on this and establish ties with the museum.

Tuesday, September 01, 2009

Facing facts

A day later the portrait of Ed Walsh is finished. Staxringold and I conominated it at the end of the evening, and overnight it became relevant at Jimbo's user talk. Here's a link to the ongoing discussion as of this writing.

An editor has found something objectionable about an attempted restoration (not one of mine, and not a featured picture) and followed up a complaint to the photography wikiproject with a complaint to Jimbo. The first I heard about this was a ping from a friend while I brewed the morning coffee. Jimbo had already brought my name into discussion and his description needed followup. That context and followup have been posted to his user page; another thought deserves space at the blog.

Occasionally someone who doesn't do restoration at all comes along and attempts to dictate parameters that wouldn't work. This doesn't occur very often, and it's been a while since the last episode. In this particular instance the editor wants to restrict the editing of historic images to levels adjustment in order to prevent what that editor perceives as original research. It's worth mentioning here that the Ed Walsh example is particularly telling in that regard.

Despite all the time that went into scratch removal, the most significant change between the original and the restoration is correction for the overexposure on parts of his face. That correction was done by using the shadow/highlight tool on a selection of Walsh's face and collar. This is a sophisticated type of levels adjustment and if I had wanted to misuse it the result would have made him look like a wrinkled old man. Compared to the histogram work (which didn't take long but was actually quite delicate) all the preceding hours of dirt, scratch, and stain removal amounted to very conservative editing. Walsh already has bags underneath his deep set eyes. A bad histogram adjustment could make him into a zombie.

Back when Johann Wolfgang von Goethe completed the second part of Faust, a publisher objected to some of the language in the Walpurgisnacht scene. Faust obligingly edited per the instructions although there was nothing all that offensive about the particular words; it was likely the pagan context that might disturb devout Christians. A later publisher commented that the Walpurgisnacht scene actually becomes less offensive with the original wording restored. You see, Goethe not only did as he was told; he also censored things that hadn't been part of the publisher's instruction. So "The witch farts" became "The witch f---s", which carries exactly the same more obscene implication in German as it does in English.

That can be the result when an artist is compelled to operate within restrictions which he regards as dunderheaded.