Active Learning for Multi-Label Image Annotation

Active learning is useful in situations where labeled data is scarce, unlabeled data is available and labeling has some cost associated with it. In such situations active learning helps by identifying a minimal set of items to label that will allow the training of an effective classifier. Thus active learning is appropriate for annotation tasks in multimedia, particularly in image labeling. In this paper we address the challenge of using active learning for multi-labeling of images in personal image collections. Multi-label learning covers situations where objects can have more than one class label and a learner is trained to assign multiple labels simultaneously. In this paper we report results on a learning system for labeling personal image collections that is both active and multi-label. The focus of the research has been to reduce the overall number of images that are presented to the user for labeling.