A Fast Face Segmentation Based on Color Spatial Features and K-means Clustering Ensembles

In this paper a fast and robust face segmentation method is presented for various face sizes in an image. The method applies the skin color features extracted in the color spaces and a k-means clustering ensembles. There are three stages are included in the proposed method. The first, the skin-color pixel feature vector included both its position and color information is extracted. For providing fast and stable classification, a k-means clustering ensembles approach which combine the clustering results obtaining from a set of k-means clusters started from a random initialization is employed. The consensus partitions of skin color pixel feature vector can be obtained based on voting mechanism. By taking face region property into account, the optimum region boundaries are then obtained by frame integration and frame segmentation algorithms those are used for merging frames and partitioning different faces in the same region respectively. Finally, candidate face regions will be found by rejecting the framed regions when its ratio of height to width is over than 2.3. The face verification of these candidate face regions can be effectively achieved by performing an appearance-based method with spectral histograms as representation and support vector machines (SVMs) as classifiers.