Semi-supervised learning of emblematic gestures

This study describes a method for semi-supervised learning of three-dimensional emblematic gestures. Starting from a supervised learning stage using a small initial training set, the training set is extended fully automatically by employing a semi-supervised learning approach. Several criteria for acceptance or rejection of the class labels generated by the classifier are proposed. The experimental evaluation shows that the proposed semi-supervised learning approach yields an error rate which is only slightly higher than that of a classifier using all manually assigned class labels.