Objectives
The first main objective of this second edition of GMPRDIA tutorial is to introduce the graph-based structural pattern recognition to the students, postdocs, experienced researchers and practitioners.
The second main objective is to present a survey of the related works in order to motivate/facilitate them to benefit from the graph-based structural pattern recognition approaches. For presenting a topic in our tutorial, we will follow the approach of first presenting the core ideas, then some example applications from literature and finally the pointers to the related works from literature.
The following ordered list presents the list of objectives of the proposed second edition of GMPRDIA tutorial:
- Introduce students and newcomers to major topics of Document Analysis and Recognition (DAR) research
- Survey a mature area of DAR research and/or practice
- Motivate and explain an DAR topic of emerging importance
- Provide instruction on established practices and methodologies
- Introduce expert non-specialists to a DAR subarea
Motivation
Graph-based representations have been used with considerable success for solving many problems in Computer Vision, Pattern Recognition and (Document) Image Analysis. This is due to its ability to model invariance for object shape and image (or scene) structure even in the presence of transformations, noise, distortions etc. The general principle to work with graphs consists of modeling local features as the nodes and their relationships (interactions) as the edges and then achieving a specific task (recognition, detection etc.) with a form of graph matching algorithm. Therefore, having a good grasp on graph-based representations and methodologies is absolutely necessary in the Pattern Recognition community.
Furthermore, in the recent years, graph neural network has caught a lot of attention of the research community. These related methods have been successfully used for the classification, labeling of graphs coming from various domains. In this scenario, obtaining desired knowledge on graph neural network and related topics is absolutely important for the community.
In this tutorial, we will present different graph representation paradigms suitable for Computer Vision, Pattern Recognition, and Image analysis research. Apart from that, different state-of-the-art techniques including graph matching, graph embedding/kernel, graph neural network for graph-based recognition, detection and indexing will be discussed and their application to the related fields will be described.