The role of the tutorials is to provide a platform for a more intensive scientific exchange amongst researchers interested in a particular topic and as a meeting point for the community. Tutorials complement the depth-oriented technical sessions by providing participants with broad overviews of emerging fields. A tutorial can be scheduled for 1.5 or 3 hours.
Symbolic and Structural Models for Image Understanding
Jamal Atif is Professor of Computer Sciences at Université Paris-Dauphine and member LAMSADE (a CNRS/Paris Dauphine mixed research unit). From 2010 to 2014, he was associate professor at Université Paris-Sud 11, member of the Machine Learning and Optimisation Team at LRI (INRIA/CNRS/Paris-Sud mixed research unit). From 2006 to 2010, he was a research scientist at IRD (Institut de Recherche pour le Développement), Unité ESPACE S140 and associate professor of computer sciences at the University of French West Indies. He received a master degree and a PhD in computer sciences and medical imaging from Université Paris-Sud in 2000 and 2004. His research interests focus on AI-based methods for image understanding, knowledge representation and reasoning, machine learning and brain imaging. He works on fields arising from mathematical morphology, graph theory, uncertainty management, logics and information theory. He was the co-advisors of three PhD students and is advising or co-advising five PhD students.
Télécom ParisTech, Université Paris-Saclay
Isabelle Bloch is graduated from the Ecole des Mines de Paris, Paris, France, in 1986, she received the Master's degree from the University Paris 12, Paris, in 1987, the Ph.D. degree from the Ecole Nationale Supérieure des Télécommunications (Telecom ParisTech), Paris, in 1990, and the Habilitation degree from the University Paris 5, Paris, in 1995. She is currently a Professor with the Image Processing and Understanding Group, LTCI, Telecom ParisTech. Her research interests include 3D image and object understanding, computer vision, artificial intelligence, 3D and fuzzy mathematical morphology, information fusion, fuzzy set theory, structural, graph-based, and knowledge-based object recognition, spatial knowledge representation and reasoning, and medical imaging.
Abstract: In this tutorial, at the cross-road of artificial intelligence (AI) and visual information understanding, symbolic and structural models for high level image and scene understanding will be presented, including knowledge representation and reasoning methods. Examples in medical and satellite imaging will illustrate these methods.
Keywords: Knowledge representation, graphs, hypergraphs, ontologies, conceptual graphs, reasoning, image and scene understanding, semantic gap, inexact graph matching, constraint satisfaction problems, logic-based reasoning for ontologies, grammars, qualitative spatial reasoning.
Aims and Learning Objectives: Image understanding is a privileged domain for the development of AI methods. In particular, models are gaining importance in image processing and understanding. While models such as numerical and statistical ones and models generally related to low level information are widely used, symbolic and structural models are more and more developed recently. It seems therefore timely to provide the ICPRAM audience with an overview of the recent advances on the models and reasoning methods at the cross-road of AI and image understanding. Besides the statistical paradigm, exploiting large image resources and databases, the symbolic paradigm, exploiting expert and domain knowledge to guide data (such as images) analysis and understanding, is of prime importance, and deserves to be better known and further developed. Researchers and engineers will find useful information for dealing with image problems for which knowledge is available and can be efficiently expressed in a structural and symbolic way. Examples include medical and remote sensing image understanding from real-world applications.
Knowledge representation, graphs, hypergraphs, ontologies, conceptual graphs, reasoning, image and scene understanding, semantic gap, inexact graph matching, constraint satisfaction problems, logic-based reasoning for ontologies, grammars, qualitative spatial reasoning.
Researchers in image analysis and pattern recognition interested in understanding how formal approaches stemming from AI can be applied to their field of interest.
What is image interpretation?
A few examples
A few methods: data vs. knowledge driven
Focus: knowledge-based approaches
2. Representations of spatial information:
Spatial entities and spatial relations
Qualitative / symbolic representations
3. Ontologies, graphs, grammars and constraint satisfaction problems:
Ontologies and description logics for image interpretation
Graphs: representations of spatial entities and spatial relations, graph-based reasoning
Conceptual graphs, constraint satisfaction problems, applications in scene understanding
Stochastic grammars and graph parsing
4. Conclusion and discussion