Abstract: Image annotation has recently received much attention as a result of the rapid growth in image data. Several works have been proposed on AIA, especially, in the probabilistic modeling and classification-based methods. This study presents a review of the image annotation methods which has been published in the last 20 years. Emphasis is mainly on the machine learning models and the classification of the AIA methods into 5 categories of decision tree-based, Support Vector Machine (SVM)-based, k-Nearest Neighbor (kNN)-based, Deep Neural Network (DNN)-based and Bayesian-based AIAs. A comparison of the five types of AIA approaches was presented based on the underlying idea, feature extraction method, annotation accuracy, computational complexity and datasets. Furthermore, a review and explanation of the evaluation metrics used were presented. Emphasis was also placed on the to carefully consider these aspects during the development of new techniques and datasets for future image annotation tasks.