Aims and Scope
As computer power increases, researchers use more complex tools for Pattern Recognition and Image Analysis. Graph theory is such a framework. Along the overall processes from a stimulus to its interpretation, graphs are used for several distinct tasks: hierarchical graphs for image segmentation and for control of perceptual strategies, graph matching for recognition and image understanding, graph manipulation for clustering, conceptual graphs for representation of relational and structural knowledge, even involving time explicitely represented within the graphs.
For instance, many works have already been done over the past few years showing that graphs are very well suited for image segmentation and more generally for image analysis. Graphs are efficient as a processing and representational scheme in pattern recognition and image processing when complex and irregularly sampled data need to be synthesized.
The goal of this Technical Committee is to federate and to
encourage research works in Pattern Recognition and Image
Analysis within the graph theory framework. Among the topics
TC-15 intends to cover, we can list the following (note that
some of these topics are related to other domains like obviously
graph theory, automata theory, finite state machines, robotics,
automation and control, data flow diagrams/ Petri nets) :
- graph based clustering ;
- graph matching ;
- graph based image segmentation ;
- irregular (graph) pyramids ;
- graph representation of shape ;
- graphs vs voronoi diagrams ;
- graphs vs rigid structures (quadtrees, pyramids) ;
- aspect graphs ;
- graph transformations ;
- graph navigation.
- graph embedding and graph kernels
The TC15 is animated by a Chair and a Cochair with the support of a steering committee. Some members are also responsible of specific tasks. The current organization is as follows:
- Pasquale Foggia
- Vice Chairs:
- Luc Brun
- Bin Luo
- Steering Committee:
- Walter Kropatsch
- Edwin Hancock
- Mario Vento
- Xiaoyi Jiang
- Graph based algorithms benchmarking (Support and promotion):
- Mario Vento
- TC15 News Letter:
- Jean-Yves Ramel
- Donatello Conte
- TC15 Web site:
- Luc Brun
We agree to focus on some special "challenges" which can help our community to federate its research. These challenges are:
- Strategies for matching graphs having large set of nodes
- Increasing the "intelligence" of a pixel-based graph
- n-D representations
Very powerful tools already exist for matching graphs. However, due to their complexity, these algorithms cannot be used for graphs having very large sets of nodes (say more than 1000). Because we cannot change these tools, we can work on the "data", i.e. the graphs. The goal of this "challenge" is to propose new strategies involving, for instance, graph reduction, stochastic or heuristic sampling, and other techniques aimed at reducing the computational cost.
When working with image based features, we often end up with a RAG-like graph which is very useful for image analysis but not for image and pattern recogntion, because it takes only into account very low level relations. How can we enhance such a graph without loosing of course its useful properties?
This topic is of great importance if we want to take into account the mistakes that have been made in the past within the research on edge detection (some authors proposed many new optimal edge detectors only because we had no tools to make them compete, or just to compare and classify them, on the basis of real data).
Graphs can be used for image feature representation. In the past this usage has mainly regarded representation in a 2D space, but in recent years some interesting works on 3D representations have been proposed. What about 4D? Are graphs a useful representation for temporal data?
Our future activities will be centered on the following main themes:
- Enhancing the divulgation of the scientific results of our community
- Increasing the interchanges with other groups/communities
- Stimulating the diffusion of the techniques developed by our community into applicative contexts
In order to make more easy the validation and the reproduction of our results, and thus to make them accepted as more significant, it is important to continue the definition of standard databases to be used as test cases (making different approaches directly comparable), and to develop benchmarking protocols for the different kinds of graph-related problems that are of interest for the TC15 members.
In order to favor the interchange with other communities the idea is the inclusion of special sessions related to graph-based techniques in conferences and workshops targeted at a wider audience, and the organization of joint events with other groups.
The individuation of relevant applications can help our research activity to remain focused on important issues; moreover, a demonstration of the impact of our methods on practical applications can attract the interest of a wider group of researchers, in the industry as well as in the academy.