Use of Semantic Enhancements to NLP of Image Captions to Aid Image Retrieval

This paper proposes a semantic-based create and search technique to enhance visual information retrieval. Our approach includes an ontology-based scheme for the semi-automatic annotation for image retrieval. Latent Semantic Indexing (LSI) is used in order to solve the Natural Language (NL) vagueness problem and to tolerate ontology imperfections. In addition, our framework is able to find indirect relevant concepts in images and to represent image semantics at a higher level. Experiments demonstrate that semantic-based approaches can significantly improve image retrieval.