Image annotation

Abstract

This paper describes an efficient approach to image annotation. It ranked first on the recent scene categorization track of the ImagEVAL1 benchmark. We show how homogeneous global image descriptors combined with a pool of Support Vector Machines achieve very good results. We also used this approach on several well known object recognition databases to emphasize two main aspects of this research domain: the importance of contextual information in object recognition and the unsuitability of many standard databases for this task.