Web image annotation with semi-supervised feature selection

Web image annotation has become a critical research issue in recent years. It is beneficial to develop effective semisupervised feature selection methods to exploit the labeled data and unlabeled data simultaneously for web image annotation. However, the graph Laplacian based semi-supervised learning methods suffer from the fact that sparse coordinates are biased toward a constant and the Laplacian embedding often cannot preserve local topology well as we expected. In this paper we propose a novel semi-supervised feature selection method based on the second-order Hessian energy, namely Hessian Semisupervised Feature Selection (HSFS), which can overcome the drawbacks of Laplacian based methods. Extensive experiments are performed and the results show that our method is more suitable than Lapacian based methods for large-scale web image annotation.