FRE 2645 Symbol Recognition System for Graphics Documents

FRE 2645 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 ²L 3 I Laboratory, La Rochelle University, France SSPR’ 04 : 18 -20 August 2004

Introduction (2) Pre-Processing (1) (3) (6) SSPR’ 04 : 18 -20 August 2004 Connected Component Labelling Inclusion Graph Matching & XML (4) (5) Neighboring Graph Hybrid Graph

(1) Connected Component Labelling § Region aggregation algorithm § direct characteristics computation Start § Loops extraction SSPR’ 04 : 18 -20 August 2004 Step III End -3 pixels -1 pixel

(2) Pre-Processing § Connected components filtering: § Surfaces histogram analysis SSPR’ 04 : 18 -20 August 2004

(3) Inclusion Graph § Tree representation of inclusion links between components/loops § Two steps: § Labelling the components and loops images § 8 -connectivity analysis of line : (x 0; y 0) (x 0+dx; y 0) SSPR’ 04 : 18 -20 August 2004

(4) Neighboring Graph “a” § Graph representation of neighboring links between connected components. § Two steps: § Contours extension § Boundaries analysis SSPR’ 04 : 18 -20 August 2004

(4) Neighboring Graph “b” SSPR’ 04 : 18 -20 August 2004

(5) Hybrid Graph SSPR’ 04 : 18 -20 August 2004

(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 SSPR’ 04 : 18 -20 August 2004

Experiments and Results “a” § Symbol contest GREC 2003 s 1 s 5 SSPR’ 04 : 18 -20 August 2004 s 2 s 6 s 3 s 7 s 4 s 8 s 9

Experiments and Results “b” SSPR’ 04 : 18 -20 August 2004

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 SSPR’ 04 : 18 -20 August 2004