Cross-Modal Saliency Correlation for Image Annotation

摘要: Automatic image annotation is an attractive service for users and administrators of online photo sharing websites. In this paper, we propose approach exploiting the crossmodal saliency correlation including visual textual saliency. For saliency, a concept graph firstly established based on association between labels. Then semantic communities latent are detected; adopt dual-layer BoW (DL-BoW) model integrated with local features salient regions image. Experiments MIRFlickr IAPR TC-12 datasets demonstrate that proposed method outperforms other state-of-the-art approaches.