Pixel-based skin detection using sinc function

Skin detection is a major key in most human-related recognition systems. Parametric models have been proposed to effectively model the distribution of skin pixels. In this paper some limitations of these methods are reviewed and a new model based on the sinc function is proposed to overcome these limitations. Three different color spaces are chosen for experiments. The results are compared with the results of Gaussian, GMM and Elliptic boundary models. The method significantly outperforms Gaussian and Elliptic method and the results also show that the model slightly outperforms GMM with 2 Gaussians used. Steepest decent method was used for data fitting and COMPAQ database is used for training purposes.