Abstract: Automatic image annotation has emerged as a hot topic in the field of image semantic understanding due to its potential application on Web image search. To effectively access and retrieve images, a popular solution is to tag images with meaningful semantic keywords,which is considered as automatic image annotation. Various machine learning technictues have been employed extensively in the field of image analysis, and there is no exception for automatic image annotation. Existing image annotation algorithms can be roughly divided into three categories, i. c.，the classification based methods, the probabilistic modeling based methods, and the graph learning based methods, respectively. We surveyed nearly 50 key theoretical and empirical contributions in the current decade related to automatic image annotation, and discussed the spawning of related sub-fields in the process. I3y carefully analyzing what has been achieved so far,we also conjectured what the future may hold for automatic image annotation research.

Key words: Automatic image annotation, Multi-instance learning, Multi-label learning, Graph learning, Probabilistic modeling