An Adjusted gaussian skin-color model based on principal component analysis

Abstract

By combining the two standard paradigms of unsupervised learning, Principal Component Analysis (PCA) and Gaussian density estimation, this paper proposes an adjusted Gaussian skin-color model for skin-color detection. This method is more robust than the standard Gaussian model because it can weaken the bias caused by noise and enhance the fitness of the mathematical model. The experiments show that this method works well for the real-world images with complex backgrounds. © Springer-Verlag 2004.