When the frescoed ceiling of Assisi's Basilica of St.
Francis crashed to the floor two years ago few people would have imagined that the resulting fragments of painted plaster would present information technology with one of its most intriguing challenges.
The earthquake of Sept. 26, 1997, killed four people in the church and transformed priceless paintings by Giotto and Cimabue into a heap of dust and colored rubble. Since then, restorers have been painstakingly piecing together the fragments and have enlisted the help of computers to tackle their most difficult task: the reassembly of Cimabue's priceless 13th century portrait of St. Mathew.
The triangular 32 square-meter painting, which shows the evangelist reading at a desk, was shattered when it plunged 22 meters to the floor. "The painting was so badly damaged that its reassembly would probably be impossible without the assistance of computer technology," said Professor Giovanni Iacovitti, who is coordinating the IT effort.
Iacovitti, head of the Information and Communication Department at Rome's La Sapienza University, leads a team of IT experts from the university and art restorers from the Central Institute of Restoration (ICR) in a joint project seeking a "virtual" solution to the brain-teazing St. Mathew puzzle. Fundedby the Italian government and the European Union, the project is about to move from an experimental stage to tackling the reassembly of the Cimabue original.
Initial results are encouraging. The team began by creating a copy of part of the fresco, measuring 40 centimeters by 40 centimeters, and breaking it into fragments similar in size to those of the original -- Cimabue's work is currently broken into around 120,000 fragments with an average diameter of 2 centimeters. Laid out in trays on a black foam base, the pieces were digitally scanned by a 6,000 x 7,000 pixel camera belonging to the ICR.
Using C++ and Mathlab software on Pentium workstations, the team developed its own software for the isolation and acquisition of the fragments and went on to develop a program that matches the pieces with the corresponding parts of a photographic reproduction of the original.
"We decided to develop algorithms of our own because we found the commercially available programs to be unsatisfactory," said Iacovitti. "They were unable to cope with ambiguity, for example where an area of black from the fresco met the black foam of the background."
The programs must adjust the scale to that of the available photographs and harmonize the colors of the photos with the results of the digital scanning.
They also have to compensate for geometric distortion, since the original work was attached to the curved surface of the basilica's vault. Finally, theymatch the fragment images with their equivalent in the unfragmented fresco -- each examined fragment rotating constantly until it finds its location in the original.
"In the laboratory experiment the computer succeeded in matching around half of the fragments, but the part where it was successful is the most important part of the painting. The computer can't cope with fragments of a uniform color: it only comes into its own where there is pictorial detail," Iacovitti said.
Giotto paintings, already manually reassembled in an Assisi workshop, were pieced together by restorers on top of scale photographs of the originals. In the case of the Cimabue, there is no scale photograph available, but restorers will have the computer's helpful suggestions to guide them. Professor Iacovitti's program provides likelihood maps, sometimes suggesting a single match, sometimes offering a range of choices listed in order of probability.
The computer cannot compete with the visual recognition capabilities of humans, but it performs a repetitive task tirelessly, accurately and fast.
Restorers will begin applying the technique to the original fragments next spring and hope to finish by the end of the year. Digital scanning was completed last March, with around 900 trays of fragments creating a 150G-byte database and filling 130 CDs. Each fragment is individually memorized and has its contours highlighted, while its individual code tells restorers where to find it.
Professor Iacovitti believes his team has developed the world's most advanced image processing technology. "I don't know of any other laboratory that has a more advanced technique. If I did, I would have used it," he said. "The mathematical techniques we have used for Cimabue can be applied to other domains of content-based image classification and to image browsing," he said.
There have also been valuable lessons for the future of art conservation. "We now know at what precision art works should be photographed for our techniques to be most effective."
The final result is unlikely to do justice to Cimabue. Parts of his fresco were irremediably destroyed by the violent impact with the floor and the colors had deteriorated with age. But whatever visual impression St Mathew makes, his intellectual challenge has brought a great leap forward forimaging technology and computer-aided art restoration.