Image Auto-annotation using 'Easy' and 'More Challenging' Training Sets
Jiayu Tang, Paul H. Lewis · ePrints Soton (University of Southampton) · 2006
Abstract The Corel Image set [1] is widely used for image annotation performance evaluation although it has been claimed [2] that the set is easy to annotate. The aim of this paper is to demonstrate some of the disadvantages of sets like the Corel set for effective auto-annotation evaluation. We first compare the performanace of several annoatation algorithms using the Corel set and find that simple near neighbour prop-agation techniques perform almost as well as the best of the more sophisticated algorithms. We then build a new image collection using the Yahoo Image Search engine1 and query-by-single-word searches to create a more chal-lenging annotated set automatically. Then, using two very different image annotation methods, we demon-strate some of the problems of annotation using the Corel set compared with the Yahoo based training set. In both cases the training sets are used to create a set of annotations for the Corel test set. Finally we show how self-annotation can be used to improve the original annotations of our Yahoo set. 1