Probabilistic collaborative representation based orthogonal discriminative projection for image set classification

Image set based collaborative representation and classification(ISCRC) has been proposed and achieved state-of-the-art performance. Though ISCRC works well for Image set based face recognition(ISFR), the classification mechanism of ISCRC is still unclear. Besides, another challenge that ISCRC encountered is to deal with the high-dimensional data. In this paper, we first propose a novel Probabilistic Collaborative Representation based Classifier for Image Set (ProCRCIS), which is interpreted from a probabilistic viewpoint. Then, according to the reconstruction residual-based classification rule of ProCRCIS, we propose a novel dimensionality reduction method, called Probabilistic Collaborative Representation based Orthogonal Discriminative Projection for Image Set(ProCR-ODP-IS). The goal of ProCR-ODP-IS is to find a projection space such that the between-class reconstruction residual is maximized and the within-class reconstruction residual is minimized simultaneously. Hence, this projected space can fit ProCRCIS very well. Extensive experimental results on different datasets demonstrate the superiority of the proposed method compared to the state-of-the-arts.