QGAR : Querying Graphics through Analysis and Recognition (project-team)

About the project : Project-Team site Activity report Research reports Theses Theme : Cog Cognitive systems Nancy - Grand Est research center

Project-Team Presentation

LORIA joint project-team with CNRS, University Henri Poincaré, University Nancy 2 and INPL.

The main scientific domain of the QGAR project-team is that of graphics recognition, i.e. the analysis of graphics-rich documents. The objectives are indexing and information retrieval, in the context of technical documentation. The basic problem is that of converting a weakly structured information, such as the image of a paper document, or a PDF file, for instance, to richer information, with structures for taking advantage of the document within an information system.

Research themes

• Feature extraction and segmentation: this has been at the center of our research for several years, as it is fundamental for a complete graphics recognition system to master the robustness, precision and reliability of the low level features (vectors, components...) extracted from a scanned document. The main scientific activity is on the algorithms for segmentation, and on the statistical models for robust estimation of vector features and for performance evaluation.
• Charaterization and recognition of signatures and symbols: symbol recognition consists in localizing and identifying the symbols present in a graphics document. Whereas many pattern recognition methods require a preliminary segmentation step, to extract regions which are candidates for recognition as a symbol, our aim is to design methods which are robust to noise, efficient and as generic as possible, for recognizing symbols without prior segmentation. Also, as direct symbol recognition is often a very hard problem, we investigate the possible use of signatures, to easily find fundamental geometric properties.

International and industrial relations

• France Télécom R&D : RNTL Docmining, doctoral thesis (CIFRE) on symbol recognition
• Algo'tech : European project on symbol recognition in electrical drawings
• ACI Madonne on legacy documents from the cultural heritage (L3i la Rochelle, PSI Rouen, LI Tours, LIRIS Lyon, IRISA)
• Collaboration with CVC University of Barcelona on signature and symbol recognition
• Collaboration with City University of Hong Kong on performance evaluation in graphics recognition

Software

Qgar Software

Scientific leader

Salvatore-Antoine TABBONE     [homepage]
+33 3 83 59 30 39
Antoine.Tabbone@inria.fr
Secretary : +33 3 83 59 20 72