A lot of computer vision applications benefit from robust skin color classification. This is due to the various image conditions like camera settings, illumination, light source, shadows and many more. Furthermore people’s tans and ethnic groups also extend those conditions. In this context, a parametric skin color classifier that can be adapted to the conditions of each image or image sequence. This is done by evaluating some previously know skin color pixels which are acquired by applying a face detector. This approach can distinguish skin color from very similar color like lip color or eye brow color. Its high speed and high accuracy makes it appropriate for real time applications such as face tracking and mimic recognition.
Skin color classifiers often specify some bounds that narrow down the color space (e.g. RGB) to the subspace of skin color. In order to gain resistance towards various illumination conditions such as the color of the light source and shadows a different color space is used. The chromatic color space (also known as “normalized RGB”) uses the proportional part of each color:
BASE = R+G+B
Rc = R/BASE
Bc = B/BASE
Rc + Gc + Bc = 1
This color space provides a commonly known way to quickly detect skin color pixels via the following classifier:
Skin_color= ( Rc > 0.35 ) ^ (Rc < 0.5) ^ (Gc > 0.2) ^ ( Gc < 0.7) ^ (BASE > 200)