A unified approach for skin colour segmentation using generic bivariate Pearson mixture model

Skin colour segmentation is rapidly growing area of research in computer science for identification and authentication of persons. In this paper, a novel generic bivariate Pearsonian mixture model for skin colour segmentation is proposed. It is observed that the hue and saturation of the colour image better characterise the features of the individual human races. In general, the human race can be characterised in to three categories namely Asian, African and European. The feature of the skin colour of these races are modelled by three different bivariate Pearsonian distributions. The combination of all these three races of people in an image can be characterised by a three component mixture model. Deriving the updated equations of the EM-algorithm, the generic bivariate Pearson mixture model parameters are estimated. The initialisation of the model parameters are done through moment method of estimation and K-means algorithm. The segmentation algorithm is developed using component maximum likelihood under Bayesian frame. The performance of the proposed algorithm is carried by experimentation with random sample of five images and computing the segmentation performance metrics such as PRI, GCE and VOI. The efficiency of the proposed model with that of bivariate GMM is carried through confusion matrix and ROC curves. It is observed that the proposed algorithm outperforms the existing algorithms.