Multi-View Semi-Supervised Learning Based Image Annotation

摘要: This paper proposes a novel multi-view semi-supervised learning scheme to improve the performance of image annotation by using multiple views an and leveraging information contained in pseudo-labeled images. In training process, labeled images are first adopted train view-specific classifiers independently uncorrelated sufficient views, each classifier is then iteratively re-trained initial samples additional based on measure confidence. unlabeled assigned appropriate semantic annotations maximum vote entropy principle correlationship between with respect results optimally trained classifier. Experimental general-purpose database demonstrate effectiveness efficiency proposed scheme.