Stanford Digital Forma Urbis Romae Project
Fragment Color Image Processing
The digital color images of the fragments
have undergone a number of processing steps from the original
photography to the images presented in the database:
- Original photograph. The fragments were photographed with
a Sony DKC-ST5 digital camera, which produces 24-bit color
images at 1030x1300 resolution.
- Lens distortion correction. The raw images were processed
with custom software which corrects for two types of lens distortion:
Because the lens distortion correction program requires resampling
of the image, minor image degradation (blurring) can occur during
- Lateral chromatic aberrations.
aberration correction spatially shifts the red and blue image channels
separately by predetermined offsets and linear scaling. These shifts
were on the order of one half pixel distance.
- Geometric distortion. The geometric distortion
correction warps the image to primarily address a pincushion-style
distortion which was present in the original images; some points
were shifted a distance on the order of several pixels.
- Mosaicing. 227 of the fragments were photographed with
multiple images that required mosaicing together (these images
are labeled as such in the database). Most of the
mosaics were made from only two or three source images and were
interactively stitched together using Panavue's ImageAssembler
software. This software reads the source images, performs
registration, color equaliziation, and resampling of the images,
and outputs a single large mosaiced image.
44 of the fragments could not be stitched adequately using this
software, and were instead completed using PTGui, a
graphical interface for the PanoTools software suite
for creating panoramas and stitching images. This software works
similarly to ImageAssembler, but allows for many more control points
in the stitching process and more flexibility in anchoring the
adjustments made for alignment and color and brightness equalization.
Mosaiced images may exhibit parallax artifacts due to
the image being a composite of several different camera
positions. These artifacts are usually manifested as blurring
along the edges of the fragments, and were removed manually
with an image painting program.
Other minor mosiacing artifacts may be visible along the boundaries
where sub-images were stitched together, as seen in the example
below (Fig. 1). The image color, contrast, sharpness, or
geometic alignment may vary slightly across the interface.
- Rotation. Images were rotated so that the fragment orientation
generally matches the orientation in the PM1960 plates, when possible.
Note that rotation may require another image resampling step, and
thus very slightly reduce the image quality.
- Cropping. The images were cropped so as to have a semi-uniform
width rectangular black border around the edges of the fragments.
- Background Masking. The fragments were photographed on
a black felt background, and a greycard and blue plastic ruler
was included in most frames. To remove these background objects
and other debris from
the images, the fragment shapes are manually traced to create
masks or mattes which correspond only to the
foreground shape of the fragment. The portions of the background
outside the mask is colored black.
- Color Correction. Most of the original images are too dark for
reasonable display, and required lightening via color remapping.
Because the fragments were not photographed under consistent
lighting and camera conditions, the specific color correction steps
applied may vary depending on the image. Typically, the color
correction involved a brightening by uniformly increasing the
midtone color intensity using the "Color Balance" function in
- Compression. The images are lossily compressed for web
distribution in JPEG format.
Other possible artifacts remaining in the images:
Figure 1: Variation between two subimages visible across the
the stitching boundary in a mosaic.
- Scaling. The images were originally photographed at
an approximate resolution of 100 pixels per inch, and are distributed
in the database at an approximate resolution of 50 dpi.
However, the precise resolution varied during the photography
process, so the
images do not have consistent scale and vary
approximately 10% about the mean.
- Color Accuracy. The colors of the fragments depicted
in the images do not necessarily accurately reflect the actual
David Koller & Kevin Coletta