Symbol Recognition System for Graphics Documents Combining Global Structural Approaches and Using a XML Representation of Data

1
Symbol Recognition System for Graphics Documents
Combining Global Structural Approaches and
Using a XML Representation of Data

• Mathieu Delalandre¹, Eric Trupin¹, Jean-Marc
  Ogier²
• ¹PSI Laboratory, Rouen University, France
• ²L3I Laboratory, La Rochelle University, France

2
Introduction
3
(1) Connected Component Labelling

• Region aggregation algorithm
• ? direct characteristics computation
• Loops extraction

Step I -3 pixels
Step II -3 pixels
Start
Step III -3 pixels
End -1 pixel
4
(2) Pre-Processing

• Connected components filtering
• ? Surfaces histogram analysis

5
(3) Inclusion Graph

• Tree representation of inclusion links between
  components/loops
• Two steps
• Labelling the components and loops images
• 8-connectivity analysis of line (x0y0)
  (x0dxy0)

6
(4) Neighboring Graph a

• Graph representation of neighboring links between
  connected components.
• Two steps
• Contours extension
• Boundaries analysis

7
(4) Neighboring Graph b
8
(5) Hybrid Graph
9
(6) Graph Matching XML

• Inexact graph-matching algorithm
• ? similarity criterion overlap between candidate
  and model graphs
• Graph editor
• ? open java graph Base Editor
• Graph representation
• ? XGMML

10
Experiments and Results a

• Symbol contest GREC2003

11
Experiments and Results b
12
Conclusion Perspectives

• Conclusion
• Advantages
• Adaptability (combination, XML)
• Good results
• Drawbacks
• Few classes recognition problem
• Segmented symbols
• Sensitive to dilatation noise
• Perspectives
• Statistico-structural recognition
• Voronoi based approach
• Contextual pre-processing
• Grammar/graph-Matching combination