Automatic image annotation using a semi-supervised ensemble of classifiers

摘要: Automatic image annotation consists on automatically labeling images, or regions, with a pre-defined set of keywords, which are regarded as descriptors the high-level semantics image. In supervised learning, previously annotated images is required to train classifier. Annotating large quantity by hand tedious and time consuming process; so an alternative approach label manually small subset using other ones under semi-supervised approach. this paper, new ensemble classifiers, called WSA, for automatic proposed. WSA uses naive Bayes its base A these combined in cascade based AdaBoost technique. However, when training Bayesian it also considers unlabeled each stage. These classifier from previous stage, then used next The instances weighted according confidence measure their predicted probability value; while labeled error, standard AdaBoost. has been evaluated benchmark data sets, 2 sets promising results.