In this PhD project, we aim at addressing the challenging problem of high level scene understanding with a new methodological approach which integrates advantages of formal logics (description logics, formal concept analysis), and quantitative or semi-quantitative representations and reasoning in the image domain.
High level scene understanding is the task of inferring semantics from image contents in a form which is close to and suitable for application domain decision-making. This semantics cannot be considered as being included explicitly in the image itself but rather depends on prior knowledge on the domain and on the context of use of the image. Model-based image understanding is a subfield where semantics inference is guided by a model e.g. a graph or an ontology. In this context, modeling imperfections of knowledge and information is also a key aspect.