An adaptive graph model for automatic image annotation

Abstract

Automatic keyword annotation is a promising solution to enable more effective image search by using keywords. In this paper, we propose a novel automatic image annotation method based on manifold ranking learning, in which the visual and textual information are well integrated. Due to complex and unbalanced data distribution and limited prior information in practice, we design two new schemes to make manifold ranking efficient for image annotation. Firstly, we design a new scheme named the Nearest Spanning Chain (NSC) to generate an adaptive similarity graph, which is robust across data distribution and easy to implement. Secondly, the word-to-word correlations obtained from WordNet and the pairwise co-occurrence are taken into consideration to expand the annotations and prune irrelevant annotations for each image. Experiments conducted on standard Corel dataset and web image dataset demonstrate the effectiveness and efficiency of the proposed method for image annotation.