VASARI

The VASARI project was completed in March 1992. Two particular achievements of the project were two Research Laboratory systems for conservation support at the National Gallery London and the Doerner Institute Munich. They both still operate, producing valuable scientific results. A third was established at the Uffizi early in 1995. The VASARI project was regarded so positively by the EC that a number of follow-up projects were initiated and others are anticipated.

The VASARI system uses scanners to capture images directly from paintings. It uses a large stepper motor controlled scanner to move a high resolution CCD camera over the painting area to obtain patches of 3k*2k pels which are later mosaiced together to form images in CIELab format with resolutions over 10k*10k. To achieve high color accuracy, seven broad-band interference filters are used to cover the visible spectrum. A calibration technique based on a least-mean-squares fit tot the color of patches from a Macbeth Colorchecker chart gives an average color accuracy of better than 3 units in the CMC uniform color space.

To achieve a resolution of around 20k*20k for a 1 square meter painting, a positioning device is used to scan a sensor over the painting area to produce high-resolution sub frames. An area CCD camera is used. A mosaicing technique is used to join the sub frames together by calculating error in their overlapping areas.

A high-resolution commercially available camera, the Kontron ProgRes 3000, is used to give a high-resolution image for each exposure. This camera, offers a resolution of up to 3000 by 2320 pels. The sensor within the camera is a standard TV camera CCD that has been masked to reduce the size of the active sites. A Zeiss Z-Planar f4/32 lens is used because it has low distortion and a good modulation transfer function. An SBUS board  is used to connect the camera directly to the computer, giving a frame grab time of about 6 seconds per channel.

The positioning system, is used to move the camera and the light projector parallel to the plane of the painting to make a collection of sub images covering the surface of large paintings. It consists of a rigid steel base mounted on a concrete floor with vibration damping blocks with two 2.5m stainless steel rails on which the main portal moves. This portal is roughly positioned and has two vertical rods between which the horizontal axis is mounted. The camera and the light projector are attached to a platform that rests on this horizontal axis. Both the horizontal and vertical axes are motorized and can be moved under computer control.

The light projector is mounted at the rear of the camera platform and a small controller unit is used to allow the filters top to be changed under computer control. The light distribution is identical for each sub image and the same light distribution correction can be applied to each sub image. Two tungsten halogen lamps, each with a set of collimating optics and an infrared blocking filter, are used to supply light to an enclosure containing the filters. The light passes through one of seven filters in the filter box and into a second fiber optic guide which divides into six 'tails' each terminated by a frosted lens unit. These give fairly even illumination over the region.

The camera, positioning system and light projector are controlled by a Sun SPARCstation 2GS workstation. This workstation, which has a 32 MB RAM, a 24 bit graphics system and an extra KGB hard disk space, is also used to collect and process the images from the camera. A 1 GB optical disk is used to store both the raw camera images and completed calibrated images.

A set of seven broad band filters is used to cover the visible spectrum. These filters have roughly Gaussian characteristics with a bandwidth of 70nm at half the maximum transmittance. Their peak transmittances range from 400nm to 700nm in steps of 50nm, covering the visible spectrum with enough overlap so that most of the spectral information will be contained in the seven channel images recorded. Using more filters might have improved the color accuracy slightly but the associated costs are disproportionally high. The acquired data is converted to and stored in the Commission Internationale de l'Eclairage (CIE) XYZ standard color space.

During scanning, the portal is positioned by hand to give the approximate required number of pixels per millimeter. The light guides are adjusted to give the most intense and uniformly distributed illumination across the area to be imaged. The camera is roughly focused by hand and then finely focused automatically using a section of the painting illuminated with light that has maximum transmittance at 550nm. Then the automatic acquisition program ACQUIRE is started. This program executes a sequence of operations to obtain a series of overlapping sub frames covering the surface of the painting through each of the seven filters.

After acquisition is completed, the automatic calibration program CALIBRATE is started. Each image is corrected to account for non-uniformity in light distribution using the image of the white target. Seven tables, one for each filter since the precise shape of the camera response function varies with the spectral power distribution of the light reaching the sensor and with camera gain, are generated which compensate for non-linearities in the response of the CCD. A color conversion matrix is generated and is used to convert each sub image to XYZ color space.

The XYZ sub images are joined in a mosaic to make a single XYZ image of the whole painting. The program uses a manually selected common point to calculate other corresponding points and to join the sub images. Automatic mosaicing can also be done using the calculated resolution to give a starting point.

The XYZ images generated by the calibration procedure are converted to a set of RGB values to drive monitors based on their characteristics. Information about the chromaticities of the three phosphors used in the display tube, the minimum and maximum light output and the gamma of each of the electron guns is used to compute the transform. The gamut of the measured colors are uniformly reduced to fit in the smaller gamut of a CRT monitor.

Applications of the VASARI system include color change measurement, surface texture analysis and infrared reflectography.