Pixel-based Skin Detection Based on Statistical Models

Abstract

Skin detection is a preliminary step in many machine vision applications. In this paper, we propose applying the Gamma, Beta, and Laplace distributions for modelling skin color pixels in arbitrary chromaticity spaces used for parametric skin detection. Since the proposed distributions do not inherently consider the correlation between the chromaticity components, a method to eliminate the correlation between the skin chrominance information is also proposed. This enables skin modelling without concerning about the data correlation. We model the skin color pixels by applying the proposed distributions in five different color spaces. The Compaq dataset was used for evaluating the performance of the proposed method. The accuracy of skin detection on the Compaq data set was 88% and showed improvement compared to previous statistical methods