Multinomial Bayesian Kernel Logistic Discriminant Based Method for Skin Detection

This paper deals with the detection of skin pixels in color images containing different ethnic groups´ skins. These images are subject to any form of distortion such as the variation of illumination and the variety of capture devices. We have used combination scheme of linear classifiers where each one is devoted to a specific ethnic group. We used kernel Bayesian logistic regression because it outperforms many existing linear classifiers. It is shown that our scheme outperforms some existing skin detection methods that provide high classification scores.