Choice of appropriate color space: The color space representation
is often lead by the skin detection methodology
and the application. Non-parametric methods such
as histogram-based methods are not affected by the color
space representation. However, parametric modeling techniques
such as Gaussian modeling are affected by the
choice of color space. It should be noted that the parametric
models are also affected by the amount and the quality
of the training data available. The choice of color space
should also depend on the available image format and the
necessity of a particular color space in post-processing
steps. For example, some non-linear color space transformations
are too computational expensive to be used in
real-time. Dropping the intensity component so as to obtain
robust parameters against illumination conditions as
approached in many of the works actually reduces the skin
detection performance.