Converting OntoUML Diagram Images into Machine-Processable Formats to Enhance the Usability of Models

OntoUML is an important ontology language, however, most of the diagrams written in this language exist only as images within published papers, rendering them impractical for research purposes. Despite existing research on converting UML diagram images to machine-processable formats, no studies address the conversion of OntoUML images. In this paper, I present the OntoUML Image Taxonomy Extractor (OITE), which detects OntoUML diagrams using transfer learning. OITE employs image processing techniques to translate OntoUML diagram images into the OntoUML Vocabulary language. The system involves class recognition through rectangle detection and OCR techniques to extract class elements, followed by line recognition and relationship type recognition to determine class relationships. Additionally, an experiment using ChatGPT was conducted to explore the potential of using visual large language models for this task. The results are used to demonstrate the feasibility of using LLMs for simple diagram translations, compare the performance of OITE and ChatGPT, and highlight areas for further research in ontology-driven conceptual modeling.