Specialized Support Vector Machines for open-set recognition

Recently, the open-set recognition problem has received more attention by the machine learning community given that most classification problems in practice require an open-set treatment. Thus far, many classifiers were mostly developed for the closed-set scenario, i.e., the scenario of classification in which it is assumed that all test samples belong to one of the classes the classifier was trained with. In the open-set scenario, however, a test sample can belong to none of the known classes and the classifier must properly reject it by classifying as unknown. It is a common scenario in practice because unknown samples can appear at usage time. In this work, we present an extension upon the well-known Support Vector Machines (SVM) classifier, called the Specialized SVM, suitable for recognition in open-set scenarios. The proposed method bounds the region of the feature space in which a test sample would be classified as known (one of the known classes), making it finite. The same cannot be guaranteed by the traditional SVM even using the Radial Basis Function (RBF) kernel. We conducted experiments comparing the proposed method with state-of-the-art open-set classifiers and show its effectiveness.