Semantic Image Retrieval Based on Multiple-Instance Learning

Image semantics have received more and more attention because of the huge gap between the low-level features and the semantics. Semantic image retrieval is a challenging problem and became a research focus. This paper presents a semantic image retrieval approach based on multiple-instance learning. In multiple-instance learning, the training samples are bags composed of instances without labels. The proposed approach realizes the mapping from low-level features to simple semantics and the mapping from simple semantics to compound semantics by means of multiple-instance learning. The obtained semantics are used in semantic image retrieval. The experiment results show that the proposed approach is able to process semantic image retrieval more reliably and more effectively.