An efficient region-based skin color model for reliable face localization

Skin color model classification is one of the several methods used for face detection. Colorspaces which discriminate between lightness and chrominance components such as HSI, HSV, YCrCb, and so forth are commonly picked to implement skin color model as their skin detection has lightness variation independence property. However, the popular conventional single-pixel-based skin color classification based on the chrominance component might fail because of several reasons. For examples, noises and skin-color-like objects may be misinterpreted to skin and unseen skin colors may be absent which causes the fragmented skin regions. To overcome this problem, we propose a new structure of skin color model which archives maximum range of variation of each color component over a small region instead of single pixel. HSV and TSL colorspaces are chosen for this research as their lightness variation independence property. In our experiments, with our proposed face localization with region-based skin color model approach, the face detection rate is higher in comparison to single-pixel-based approaches.