Regional skin detection based on eliminating skin-like Lambertian surfaces

Skin color is an often used feature in machine vision application related to humans, regarded as pixel-based skin color detection. This method has an inherent problem when color overlaps between skin and none skin regions In this paper a skin detection system based on a regional property of objects, called; Lambertian reflection is proposed. Skin is a none-Lambertian surface. In order to use this feature, first, a Single Gaussian Model (SGM) is used to detect skin like pixels in an image. From detected skin regions, ones having none-Lambertian surface property are detected as final skin regions. To detect none-lambertian skin regions, an exponential is fit the histogram of potential skin regions and based on the fit, it is decided if the region is skin or not. By incorporating this feature the False Positive detect rate (FPR) is greatly decreased.