Semi-supervised subclass support vector data description for image and video classification

In this paper, an One-Class Classification method, namely the Semi-Supervised Subclass Support Vector Data Description, is presented. The proposed method extends Support Vector Data Description by two means, i.e. by exploiting global class information expressed by the class data variance and local neighborhood information between all available (labeled and unlabeled), following the smoothness assumption of semi-supervised learning. The derived minimum bounding hypersphere enclosing labeled examples lies in a regularized space, where the introduced properties have been expressed. The proposed method has been applied in face recognition and human action recognition problems, providing increased classification performance in comparison with related methods in both supervised and semi-supervised learning settings.