Knowledge Propagation in Collaborative Tagging for Image Retrieval

An important issue in current collaborative framework for media tagging is that some images or videos may not be annotated properly or even not annotated at all. In view of this, this paper proposes a new knowledge propagation scheme to automatically propagate keywords from a subset of annotated images to the unannotated ones. The main idea is based on image content analysis and training of keyword classifiers. An evolutionary scheme is utilized to find the salient regions in the annotated images, and the importance of the other regions is estimated using one-class support vector machine (OCSVM). An ensemble of variable-length radial basis function (VLRBF)-based classifiers is trained based on the visual features of the annotated images. The trained classifiers are then used for knowledge propagation. Experimental results using 100 concept categories demonstrate the effectiveness of the proposed method.