A Semi-supervised Orthogonal Subspace Learning with Spline Embedding

In order to improve the performance of image clustering and classification, this paper proposes a semi-supervised orthogonal projection with spline embedding (SOPSE).SOPSE utilizes both labeled and unlabeled samples to learn an orthogonal projection subspace where the separability between different classes is maximized and the separability within the same classes is minimized.At the same time, SOPSE can guarantee the manifold geometry of original high-dimensional data by transforming local coordinators to global coordinators in reduced subspace with local spline embedding.The experiments demonstrate the effectiveness of the proposed method.