A statistical and structural approach for symbol recognition, using XML modelling

1
A statistical and structural approach for symbol
recognition, using XML modelling
Mathieu Delalandre, Pierre Héroux,
Sébastien Adam, Eric Trupin,
Jean-Marc Ogier
2
Introduction

• Description System using a statistical and
  structural approach for technical symbol
  recognition with XML modelling of recognition
  results

Utility map extract
XML modelling of recognition results
3
Treatment steps
Loop extraction
(1)
Blob coloring filtering
(2)
Feature extraction
(3)
Statistical classification
(4)
XML modelling of recognition results
Image processing
Model reconstruction
(5)
Recognition processing
Structural classification
(6)
XML Modelling
4
(1) Loop extraction
5
(2) Blob coloring filtering
6
(3)(4) Feature extraction Statistical
classification
(4)Feature extraction Fourrier-Mellin
invariants Zernike moments Circular
probes (5)Statistical classifier kppv
7
(5)(6) Model reconstruction Structural
classification
(5)Model reconstruction Connection and/or
distance contraints (6)Structural
classifcation Similarity criterion based on
common sub-graph
8
Results

• Statistical results
• 50 loops for training and test set
• 99 of loop recognition
• Structural results
• 9 plan extracts, 100 loops, 30 symbols
• Without noise on statistical classification
• 100 of loop recognition gives 100 of symbol
  recognition
• With noise on statistical classification
• 55.4 of loop recognition gives 86.86 of symbol
  recognition

9
Perspectives (1)

• Exploiting sub-graph isomorphism

10
Perspectives (2)

• Exploiting structural models