Multi-graph Laplacian Feature Mapping Incorporating Tag Information for Image Annotation

Automatic image annotation has exerted a tremendous fascination of many researchers with the development of multimedia and computer vision. However, most methods employ graph learning omit to combine the label information with the manifold structure between samples or utilize the single graph when computing the sample graph. Furthermore, the visual content is ignored in the process of computing the manifold structure between labels. These drawbacks lead to incomplete and inaccurate manifold information. To this end, we propose a Multi-graph Laplacian Feature Mapping Incorporating Tag Information method for image annotation. Our method firstly combines the label information with the Laplacian eigenmaps, and multi-graphs are utilized to maintain the local geometric structure of samples. Then the visual content is taken into account to obtain tag correlations. Afterward, a sea of empirical evaluations is conducted on three benchmark datasets to prove the effectiveness of the proposed method.