Learning an Ontology for Visual Tasks

The definition of ontology for visual tasks is often very tricky, since humans are usually not so good at expressing visual knowledge. There is a gap between showing and naming. The knowledge of expressing visual experience is often not trained. Therefore, a methodology is needed of how to acquire and express visual knowledge. This methodology should become a standard for visual tasks, independent of the technical or medical discipline. In this paper we describe the problems with visual knowledge acquisition and discuss corresponding techniques. For visual classification tasks, such as a technical defect classification or a medical object classification, we propose a tool based on the repertory grid method and image-processing methods that can teach a human the vocabulary and the relationship between the objects. This knowledge will form the ontology for a visual inspection task.