Semi-supervised Discriminant Analysis for Skin Detection in Color Images

A new algorithm for skin segmentation in color images is presented in this paper based on semi-supervised discriminant analysis (SDA). At first, input RGB space input image is transferred to YCbCr and CIE Lab color space in which skin pixels are more similar to each other and different from pixels of other objects. Some components of new color spaces are treated as features of pixels and construct feature vectors. Feature vectors are given to SDA algorithm to decrease the inter-class distances and increase between-class distances. Finally, projected vectors are given to the K-nearest neighbor (KNN) classifier to separate skin pixels from non-skin pixels. Simulation results show that proposed approach has considerable efficiency in skin pixel detection.