A PFA based choice of color components based on skin detection approach

Detection of human skin in colored images has always been performed in known standard color spaces. In this paper a new color space coordinate is proposed based on popular existing color spaces but taking into account the most representative ones. Color components are considered features from which a representative set is derived using the Principal Feature Analysis (PFA). An elliptical model based classifier is used to test the performance of the proposed space transformation. A Markov Random Field is used to model the relations among neighboring pixels to further assign pixels to their final clases.