Improved skin detection based on dynamic threshold using multi-colour space

Skin colour detection is widely used in applications such as adult image filtering, steganography, content-based image retrieval (CBIR), face tracking, face recognition, and facial surgery. Recently, researchers are more interested in developing high level skin detection strategy for still images based on online sample learning approach which requires no offline training dataset. Previous dynamic skin color detection works has shown high true positive result than the static skin detection in term of skin-like colour and ethnicity factors. However, dynamic skin colour detection also produced high false positives result which lowers the accuracy of skin detection. This is due to the current approach of elliptical mask model that is not flexible for face rotation and is based on single colour space. Therefore, we propose dynamic skin colour detection based on multi-colour space. The result shows the effectiveness of the proposed method by reducing the false positive rate from 19.6069% to 6.9887% and increased the precision rate from 81.27% to 91.49%.