Fast grid-based clustering method for automatic calculation of optimal parameters of skin color classifier for head tracking

One of the fast method for extracting head position from the camera image is the selection of face pixels with a skin color classifier. Regardless of the chosen model of such a classifier, the problem is to automatically calculate its optimal parameters and to obtain a short computation time. In this article a new grid-based method for clustering RGB space is proposed. It allows a significant reduction in large data set size with an accurate representation and a short calculation time. The clustering method is multipurpose as it does not require any input parameters, and can operate on the large data sets and makes no assumptions about clustered data set.