Specifications

Submitted to Studies in Conservation, March 2006
3
could also use an RGB camera with and without a single colored filter [4].) By imaging a
calibration target with a number of colored patches (e.g., GretagMacbeth ColorChecker DC or
SG), a transformation is derived that converts the six signals to spectral reflectance factor. In our
research, spectral reflectance factor ranges between 380 and 730 nm in 10 nm increments with a
constant 10 nm bandwidth. When referring to this technique, the term “spectral estimation”
rather than “spectral measurement” is used because the system is mathematically
underdetermined: six input channels predict 36 output channels.
The general workflow is shown in Figure 1. We have used a Sinarback 54H digital back
with Sinar P3 body, Sinaron Digital HR 100mm lens, and Sinar m shutter in our research [5].
The camera was modified by replacing the sensor’s blue-green cover glass with a clear cover
glass. This extended the spectral sensitivity of the camera into the near infrared. A Sinar liquid
crystal shutter was modified to hold the two filters. A pair of Broncolor Pulso G2 Xenon strobes
were used to illuminate the object plane.
The camera back had micro-positioning capabilities; the sensor was moved four times to
position each color of the Bayer patterned CFA over every pixel’s spatial location. This is known
as the “four shot” or “four pop” mode and results in full color images without spatial
interpolation, but can only be used for static scenes. The four shots were repeated with each filter
in front of the lens, resulting in an “eight shot” or “eight pop” mode. Capture was controlled by a
modified version of Sinar’s CaptureShop software. For each exposure time, the software
captured an image with the shutter closed and subtracted this “dark-image” from subsequent
exposures. The software also applied a multiplicative spatial correction called the sensor shading
reference that compensated for pixel-wise differences in the camera’s gain. Both operations
were transparent from the user’s perspective. The resulting eight-plane image was stored in
Sinar’s, TIFF based, STI file format.
The remaining image processing steps were implemented using the programming
language, Matlab [6]. The next step was to re-assemble each pair of four-planes into a pair of
three-plane images. The spectral sensitivity of the Bayer-patterned CCD varied slightly for green