04.06.2005 17:47

Digital images of 'The Unicorn in Captivity': two mathematicians and a 'cluster of nodes' solve the problem


(I also blogged about this on in illo tempore at 'The Hunt of the Unicorn' tapestries' renovation reveals ....)

A 15th Century Find in Manhattan at Steve Goddard's History Wire points to Capturing the Unicorn by Richard Preston, also available in the April 11, 2005, issue of The New Yorker. From Capturing the Unicorn:
At some point, the backs of the tapestries had been covered with linen. The backings, which protect the tapestries and help to support them when they hang on a wall, were turning brown and brittle, and had to be replaced. Using tweezers and magnifying lenses, [lead conservator Kathrin] Colburn and her team delicately removed the threads that held each backing in place. As the conservators lifted the backing away, inch by inch, they felt a growing sense of awe. The backs were almost perfect mirror images of the fronts, but the colors were different. Compared with the fronts, they were unfaded: incredibly bright, rich, and deep, more subtle and natural-looking. The backs of the tapestries had, after all, been exposed to very little sunlight in five hundred years. Nobody alive at the Met, it seems, had seen them this way. ...

Philippe de Montebello, the director of the museum, declared that the Unicorn tapestries must be photographed on both sides, to preserve a record of the colors and the mirror images. Colburn and her associates would soon put new backing material on them, made of cotton sateen. Once they were rehung at the Cloisters, it might be a century or more before the true colors of the tapestries would be seen again. ...

It took two weeks to photograph the tapestries. When the job was done, every thread in every tile was crystal-clear, and the individual twisted strands that made up individual threads were often visible, too. The data for the digital images, which consisted entirely of numbers, filled more than two hundred CDs. With other, smaller works of art, Bridgers and her team had been able to load digital tiles into a computer's hard drives and memory, and then manipulate them into a complete mosaic -- into a seamless image -- using Adobe Photoshop software. But with the tapestries that simply wouldn't work. When they tried to assemble the tiles, they found that the files were too large and too complex to manage. ...
Two mathematicians, brothers Gregory and David Chudnovsky at first used a 'cluster of nodes' supercomputer they had assembled from mail order parts, in an attempt to join the tiles. But they wouldn't match: 'the warp and weft threads didn't run smoothly from one tile to the next. The differences were vast. It was as if a tapestry had not been the same object from one moment to the next as it was being photographed.'

Because the tapestry had been hanging vertically for centuries, but was photographed flat on the floor of the museum's lab, the threads began to relax, 'breath[ing], expanding, contracting, shifting. It was as if, when the conservators removed the backing, the tapestries had woken up. The threads twisted and rotated restlessly. Tiny changes in temperature and humidity in the room had caused the tapestries to shrink or expand from hour to hour, from minute to minute. The gold- and silver-wrapped threads changed shape at different speeds and in different ways from the wool and silk threads.'

How to solve the problem of calculating changes in the tapestry, in order to match the image tiles? Here the article is maddeningly terse, referring to calculations of each pixel's relationship to all nearby pixels, billions and billions of calculations. AFAICT, The New Yorker doesn't disclose Preston's email address, so surface mail or a telephone call are how to contact him for further information.

Another 'wonderful thing' a la Howard Carter and Lord Carnarvon.