Laplacian Eigenmaps Regularized Feature Mapping For Image Annotation

In the past two decades, researchers have shown great interest in automatic image annotation. However, most existing research methods do not consider similarities among samples or do not obtain the suitable manifold information. Those methods also require the adequate and precise label sets. Considering above mentioned challenges, we propose a method, called laplacian eigenmaps regularized feature mapping for image annotation, which construct a laplacian matrix with all data in the training set (include labeled data and unlabeled data) and embed the laplacian matrix into feature mapping. Experimental results conducted on several benchmark image annotation datasets, such as Corel5K and ESP Game, demonstrate the effectiveness of the proposed method.