Fast and accurate skin segmentation in color images

Abstract

We propose a new type of skin segmentation in color images especially developed for applications that require low resources. In our recent [7] work we made use of the Mahalanobis distance that proved its value in human skin segmentation. The calculation of the mentioned distance is very computational demanding. This is often a drawback to an implementation on critical applications, where speed is important. Robot applications are among this group. We now present a skin segmentation only based in each pixel color information to determine its class. We need to make an initial camera calibration and to determine the ST space that we consider skin color. This method is proved effective for most cases of human interaction. The present work can be a initial step in any application requiring human interaction or simply for skin detection.