Structural Analysis in Document Interpretation: a Primer

1
Structural Analysis in Document
Interpretationa Primer

• Mathieu Delalandre¹ , Eric Trupin¹, Jean-Marc
  Ogier²
• ¹PSI Laboratory, Rouen University, France
• ²L3I Laboratory, La Rochelle University, France
• Phd thesis website http//site.voila.fr/roxml

2
Plan

• Introduction
• Global Structural Analysis
• Local Structural Analysis
• Method Combination
• Conclusion

3
Plan

• Introduction
• Global Structural Analysis
• Local Structural Analysis
• Method Combination
• Conclusion

4
Introduction

• document interpretation
• handwriting
• structured document
• graphics (symbol, technical document, map)

5
Introduction

• shape recognition process
• statistical analysis and recognition (Jain 2000)
• structural analysis and recognition (Tombre 1996)

feature extraction or analysis (Loncarnic 1998)
recognition
ltnetworkgt ltpart id1gt ltsymbolsgt ltlabelsgtlt/labels
gt lt/symbolsgt lt/partgt lt/networkgt
6
Introduction

• structural recognition
• graph-matching approach (Hancock 2002)
• grammar approach (Blostein 1995)
• structural analysis

local
global
7
Plan

• Introduction
• Global Structural Analysis
• Local Structural Analysis
• Method Combination
• Conclusion

8
Global Structural Analysis

• applications
• structured document
• form (Altamura 2001)
• character string (Loo 2001)
• mathematical formula (Leplumey 2002)
• graphics
• technical drawing (Burge 1995)
• symbol (Delalandre 2002)

9
Global Structural Analysis

• approaches
• direct document ? object (up bottom)
• indirect connected component ? object (bottom
  up)

choice of granularity exploration as
multi-resolution analysis (Tan 2001)
10
Global Structural Analysis

• split merge (Hadjar 2001)
• graph of Voronois regions (Burge 1995)

indirect (Burge 1995)
direct (Hadjar 2001)
11
Global Structural Analysis

• method comparison
• direct methods
• advantages low complexity, high-level object
• drawbacks object segmentation
• indirect methods
• advantages object segmentation
• drawbacks low level object ? recognition step

12
Plan

• Introduction
• Global Structural Analysis
• Local Structural Analysis
• Method Combination
• Conclusion

13
Local Structural Analysis

• applications
• character
• latin (Trier 1996)
• asian (Fan 2002)
• graphics
• technical drawing (Liu 1999)
• symbol (Lladós-2001)

14
Local Structural Analysis

• four steps

object graph extraction
mathematical approximation
high level object construction
object graph correction
15
Local Structural Analysis

• object graph extraction (1)

skeletonisation
contouring
run
tracking


meshes
region
segmented object
16
Local Structural Analysis

• object graph extraction (2)
• skeletonisation based methods, two steps
• skeletonisation (Lam 1995)
• skeleton graph extraction (Tombre 1999)(Lau 2002)
• contouring based methods
• morphological based methods (Hasan 2000)
• following based methods (Ablameyko Pridmore
  2000)
• tracking based methods, two tracking types
• line tracking (Song 2002)
• junction tracking (Ogier 1992)

17
Local Structural Analysis

• object graph extraction (3)
• run decomposition based methods
• vertical and horizontal run graph (Burge 1998)
• region decomposition based methods
• orientation map (Cao 2000) and wave aggregation
  (Delalandre 2003)

18
Local Structural Analysis

• object graph extraction (4)
• meshes based methods, two steps
• vectorisation with meshes library (vaxivière
  1995)
• segmented based methods, two segmentations
• line, circle and ellipsis segmentation (Matas
  1999) (Su 2002)
• junction segmentation (Chen 2000)

19
Local Structural Analysis

• mathematical approximation
• vector, circle, and curve fitting (Rosin 1995)
• high level object construction
• circle reconstruction (Hilaire 2001)
• contour matching (Zou 2000) (Ramel 2000)

circle reconstruction
mathematical approximation
contour matching
20
Local Structural Analysis

• object graph correction, some examples
• pruning merging (Delalandre 2003)
• junction fusion (Lin 2002)
• vectorial correction (Hilaire 2001)
• and so on

pruning merging
junction fusion
vectorial correction
21
Local Structural Analysis

• method comparison (1)
• object graph extraction, seven criteria

• Junction junction detection
• Morphology various shape analysis
• Invariance scale and orientation change
• Sensitivity noise resistance
• Semantic low or high level object
• Reversibility image return
• Complexity low algorithmic complexity

22
Plan

• Introduction
• Global Structural Analysis
• Local Structural Analysis
• Method Combination
• Conclusion

23
Method Combination

• Comparison combination
• different analysis and recognition, then
  comparison
• ? multi representation
• skeletonisation and contouring (Nakajima 1999)

24
Method Combination

• hybrid combination
• different analysis, then hybrid graph
  construction and recognition
• ? multi representation
• contour concavity and region graph (El-Badawy
  2002)
• statistical structural approach (Delalandre 2003)

25
Method Combination

• cooperative combination
• first analysis in or order to help the second
  analysis
• ? process simplification
• object simplification (Song 2002)

26
Plan

• Introduction
• Global Structural Analysis
• Local Structural Analysis
• Method Combination
• Conclusion

27
Conclusion

• structural analysis
• image ? graph
• groups or decomposes the connected components
• method combination
• a research perspective
• ? multi representation
• ? process simplification

28
End

• sorry for my absent, any question ?
• Mail mathieu.delalandre_at_univ-rouen.fr
• Phd Website http//site.voila.fr/roxml