An improved approach for image annotation

摘要

The overwhelming amounts of digital images on the Web and personal computers have triggered the requirement of an effective tool to retrieve images of interest using semantic concepts. Due to the semantic gap between low-level features of image content and its high-level conceptual meaning, however, the performances of many existing automatic image annotation algorithms are not so satisfactory. In this paper, a novel approach based on the multi-view semi-supervised learning scheme is proposed to improve the quality of annotation. In the training process, labeled images are first adopted to train view-specific classifiers independently using uncorrelated and sufficient views, and each view-specific classifier is then iteratively re-trained using initial labeled samples and additional pseudo-labeled samples based on a measure of confidence. In the annotation process, each unlabeled image is assigned appropriate semantic annotations based on the maximum vote entropy principle and the correlationship between result annotations of optimally trained view-specific classifiers. Experimental results conducted on 50,000 Flickr image dataset demonstrate that the proposed scheme can effectively improve the performance of image annotation.