A novel image annotation feedback model based on internet-search

Abstract

We propose an Internet-search-based automatic image annotation feedback model, combining content-based and web-based image annotation, to solve the relevance assumption between the image and text and the limited volume of the database. In this model, we extract candidate labels from search results using web-based texts associated with the image, and then verify the final results by using Internet search results of candidate labels with content-based features. Experimental results show that this method can annotate the large-scale database with high accuracy, and achieve a 5.2% improvement on the basis of web-based automatic image annotation. © Springer-Verlag Berlin Heidelberg 2012.