Laplacian-Hessian regularization for semi-supervised classification

With exploiting a small number of labeled images and a large number of unlabeled images, semi-supervised learning has attracted centralized attention in recent years. The representative works are Laplacian and Hessian regularization methods. However, Laplacian method tends to a constant value and poor generalization in the process of classification. Although Hessian energy can properly forecast the data points beyond the range of the domain, its regularizer probably leads to useless results in the process of regression. So the Laplacian-Hessian regression method for image classification is proposed, which can both predict the data points and enhance the stability of Hessian regularizer. To evaluate the Laplacian-Hessian method, Columbia Consumer Video database is employed in the paper. Experimental results demonstrate that the proposed method perform better than Laplacian or Hessian method in the matter of classification and stability.