Open-set semi-supervised audio-visual speaker recognition using co-training LDA and sparse representation classifiers

Semi-supervised learning is attracting growing interest within the biometrics community. Almost all prior work focuses on closedset scenarios, in which samples labelled automatically are assumed to belong to an enrolled class. This is often not the case in realistic applications and thus open-set alternatives are needed. This paper proposes a new approach to open-set, semi-supervised learning based on co-training, Linear Discriminant Analysis (LDA) subspaces and Sparse Representation Classifiers (SRCs). Experiments on the standard MOBIO dataset show how the new approach can utilize automatically labelled data to augment a smaller, manually labelled dataset and thus improve the performance of an open-set audio-visual person recognition system.