Georgina Kwei, Kobus Barnard, and Brian Funt
Proceedings Colour Image Science 2000, Univ. of Derby, Derby, UK, April 2000.
Spotting an object of a chosen colour, skin for example, in an image taken under
uncontrolled lighting conditions requires some form of preprocessing to convert the
image colours to an illumination-independent specification. Computational colour
constancy algorithms are designed specifically to accomplish this task. We process
images with different colour constancy algorithms and evaluate the results in terms
of the resulting accuracy of object colour spotting. Since spotting a colour depends
on the threshold used to define the colour, ROC analysis is used in evaluating the
results. The experiments help to evaluate the performance of the colour constancy
algorithms and to predict what can be expected in terms of being able to find
particular colours in a cluttered environment.
Full text (pdf)
Keywords: colour, object recognition, computational colour constancy
Back to SFU Computational Vision Lab publications (home)