A family of skin modelling methods was designed and tuned specifically
for skin detection during face tracking. This task makes skin
detection different from the static images analysis in several aspects.
First, in principle, the skin model can be less general (more
specific) - i.e. tuned for one concrete person, camera or lighting.
Second, initialization stage is possible, when the face region is
discriminated from background by different classifier or manually.
This gives a possibility to obtain skin classification model, that is
optimal for the given conditions (person, camera, lighting, background).
Since there is no need for model generality, it is possible
to reach higher skin detection rates with low false positives with
this specific model, than with general skin color models, intended
to classify skin in totally unconstrained images set (like in [Jones
and Rehg 1999]). On the other hand, skin color distribution can
vary with time, along with lighting or camera white balance change,
so the model should be able to update itself to match the changing
conditions. Also, model training and classification time becomes
extremely important here, for the skin detection system must work
at real-time, consuming little computing power.