Birmingham Object Lighting Database
This page provides technical information and files to support to the BOLD image database
Height Maps for surfaces
Height maps for the milled surfaces OGAB1-4, OISO1 and 2
and OSINE. These
height map images were used to generate the milled surface for the surface photographs and they are registered with the photographs
except that whereas the photographs have a border these image stop at the surface edge. Pixels in the
height maps had side length = 0.045mm in the models. 1 grey level in the height maps corresponds to 0.001447mm in the z-direction of the models.
Colour space conversions
MatLab code for converting images from their LinearRGB format to sRGB, XYZ and
The photographs in the database are colour balanced for the sRGB format but have
'linear to luminance' intensity values
rather than gamma compressed values. A photograph of a ColorChecker chart
(X-rite Inc MA; similar to a MacBeth colour chart) taken with our
cameras and converted to lu'v' assuming standard illuminant A gives colour co-ordinates that are a
good representation of the ColorChecker colours under a tungsten illuminant as
measured with a SprectCAL spetroradiometer. Similarly outdoor images of the
ColorChecker card converted to lu'v' assuming a D50 illuminant produce colour
values that match those measured. Note our
cameras have not been colour calibrated but when compared to a standard they seem to produce good results.
You can download comparisons between the colour rendition from our cameras and Spectroradiometer readings for the tungsten lights
and for daylight. Note the linearity of the camreas can be verified in these spreadsheets although some saturation
occurred for the white most tile of the ColorChecker in the outdoor case. You can also download the spectral
characteristics of our light sources, including daylight.
Lighting positions and specifications for
the object, surface and face photographs. Absolute sun positions (Azimuth and
Elevation) are given for the external photographs along side the relative (ie
relative to the camera) sun positions.
Camera Geometry, toe in and registration
The centres of he camera sensors were 95.4 mm apart. For objects and surfaces the
cameras were toe'ed in by 4.2° each (8.4° total toe in). The viewing distance was was 645mm.
For the portrait images the toe in was 2.2° (4.4° total). External images had no
Image pairs were registered as follows:
Surfaces: These were registered to a common rectangle using a projective
transformation that removed trapezoidal distortions. Zero disparity corresponds
to the mean depth plane of each surface.
Objects: We assumed objects to be broadly spherical and cropped images to place
objects centrally. Thus zero disparity corresponds to the object centre.
Faces: The original image pairs varied slightly in scale (zoom) and orientation.
We corrected these errors by applying equal and opposite transformations to the
two images so as to bring the eyes into line. Equal and opposite transformations
were applied to the two images in a pair. Zero disparity is at the middle of the
nose and slighting inside the head (at eye depth) for the reference frontal pose
image but people will have moved between photographs with different poses /
Outdoor: Slight differences iorientation were corrected by aligning two
horizontal reference points between members of a pair and slight differences in
scale (zoom) were corrected based on estimate from two vertically displaced
reference points. Equal and opposite transformations were applied to the two
images in a pair. The zero disparity plane is not well defined (the camera axes
converged at infinity) but relative disparities are useful.
Page maintained by the Vision Laboratory at The
University of Birmingham all rights reserved, images on this site may be used for
academic purposes only. For information mail a.j.schofield at bham.ac.uk