A Colour Space for Skin Detection Using Principal Components Analysis Technique

Colour is one of the most important features used in skin and face detection. Colour space transformation is widely used by researchers to find better representation of human skin tone. Despite the research efforts in this area, choosing a proper colour space for skin and face detection remained an unsolved issue. Illumination variation, various camera characteristics, different skin colour tones and skin-like colours in background are among major challenges in skin detection. This paper proposes a new colour space based on projection of YCbCr colour space to principal component of three different skin colour clusters corresponding to three human ethnics including Asian, Black and Caucasian by means of a variation of principal component analysis (PCA) technique. Two classifiers including Random Forest and Support Vector Machine (SVM) have been employed to construct the skin colour model. Meanwhile, a dataset of 450 images consist of skin locus of different ethnics (Asian, Black and Caucasian) under various lighting condition was used. The proposed colour space was compared to ten state of the art colour spaces and gave superior results in term of pixel-wised skin classification performance. The experimental results show that the proposed colour space yields F-score rate of 0.9273 and ROC curve area of 0.9563 outperforms other colour spaces in this study.