Skin detection using a modified Self-Organizing Mixture Network

Skin detection is widely used in algorithms for face detection and gesture analysis. The primary step for skin detection is modeling the skin and non-skin pixels using accurate distributions. In this paper, the Self-Organizing Mixture Network (SOMN) is modified to improve its computation performance, stability and applicability, and a probability density estimation method based on the modified SOMN is put forward to establish color models for skin and non-skin classes accurately and effectively. This density estimation approach outperforms the Expectation-Maximization (EM) algorithm in various aspects such as convergence speed and estimation accuracy. According to the obtained skin and non-skin color distributions, the Bayesian decision rule is applied to classify the image pixels. Our method gives a true positive rate of 90.01% with 14.21% false positive, which is equivalent to the detection rate of the histogram model with naive Bayes classifier, and slightly better than those of the Gaussian mixture model based classifiers. And it is much superior in storage requirement and computation efficiency. Experimental results show that the proposed detection method is fast and accurate, and it can adapt to the changes in the lighting and the viewing environments.