Fast Building of Region Graph from SVG Mathieu

Fast Building of Region Graph from SVG Mathieu Delalandre, Zouba Karim, Norolala Ramangaseheno Supervisors Tony Pridmore (IPI, Nottingham University, UK) Eric Trupin (PSI, Rouen University, France) Eureka Meeting, L 3 i Laboratory, La Rochelle University Tuesday 20 th April 2006 Image Processing and Interpretation Group University of Nottingham

Introduction What are Vector Graphics ? <rect x="400" y="100" width="400“ height="200" fill="yellow" stroke="navy" stroke-width="10" /> (a) (b) Example Common formats : • AI (Adobe Illustrator) Clip. Art Cheese • SVG (Scalable Vector Graphic) • WMF (Windows Metafile) • EPS (Encapsulted Post. Script) • DXF (Auto. CAD) • Clip. Art • Flash Eureka Meeting, L 3 i Laboratory, La Rochelle University Tuesday 20 th April 2006 EPS plane WMF pen Image Processing and Interpretation Group University of Nottingham

Introduction Vector Graphics Indexing & Retrieval (1/2) • Vector graphics are growing on Web and databases • [Mong’ 03] [Chen’ 04] [Kang’ 04] … • Few I&R systems have been developed (> 2000) • [Love-01] [Sciascio’ 04] [Dosch’ 04] [Fonseca’ 05] [Rusiñol’ 05] [Zakaria’ 05] … Vector Graphics Features Extraction Matching Index Ranking Eureka Meeting, L 3 i Laboratory, La Rochelle University Tuesday 20 th April 2006 Image Processing and Interpretation Group University of Nottingham

Introduction Vector Graphics Indexing & Retrieval (2/2) • Line Graph [Dosch’ 04] [Zakaria’ 05] … • Symbol recognition • Region Graph [Fonseca’ 05] [Rusiñol’ 05]. . • Document indexing • Our works region graph extraction for SVG Indexing and Retrieval large sized data a fast approach • Two steps : unformat SVG and region graph building Eureka Meeting, L 3 i Laboratory, La Rochelle University Tuesday 20 th April 2006 Image Processing and Interpretation Group University of Nottingham

Unformat SVG Unformat process, what is it ? Example: overlapped rectangles <rect x="400" y="100" width="400" height="200" fill="blue" /> <rect x="650" y="200" width="400" height="200" fill="yellow" /> L 1 L 2 R 1 R 2 R 3 L 3 R 1 R 2 L 4 R 3 ? to unformat (or to broke) SVG Eureka Meeting, L 3 i Laboratory, La Rochelle University Tuesday 20 th April 2006 Image Processing and Interpretation Group University of Nottingham

Unformat SVG Overview of our approach SVG document Parsing for line extraction filtering set of (joined) lines Eureka Meeting, L 3 i Laboratory, La Rochelle University Tuesday 20 th April 2006 set of no joined lines Image Processing and Interpretation Group University of Nottingham

Unformat SVG Why using a filtering step you see 5 lines Eureka Meeting, L 3 i Laboratory, La Rochelle University Tuesday 20 th April 2006 you have 9 lines Image Processing and Interpretation Group University of Nottingham

Unformat SVG Our filtering process l 1 same as l 2 (a) l 1 intersects and overlaps l 2 (a) (b) l 1 includes l 2 l 1 intersects only l 2 (a) (b) l 1 joins l 2 Eureka Meeting, L 3 i Laboratory, La Rochelle University Tuesday 20 th April 2006 Image Processing and Interpretation Group University of Nottingham

Unformat SVG Examples of results crossing point merged lines Eureka Meeting, L 3 i Laboratory, La Rochelle University Tuesday 20 th April 2006 Image Processing and Interpretation Group University of Nottingham

Region Graph Building How it works ? K Lines Line Graph Finding regions and their links Eureka Meeting, L 3 i Laboratory, La Rochelle University Tuesday 20 th April 2006 Image Processing and Interpretation Group University of Nottingham

Region Graph Building Approaches used in the literature • Approaches based on graph handling [Fonseca’ 05]. . • region detection = problem of finding minimum length cycles inside a graph Eureka Meeting, L 3 i Laboratory, La Rochelle University Tuesday 20 th April 2006 Image Processing and Interpretation Group University of Nottingham

Region Graph Building Our approach (1/3) • Based on [Weindorf’ 01] works : using vectorial information Definition: direct angle = anticlockwise α(2 -1) L 1 α(1 -2) L 4 L 5 L 2 L 3 Vector Graphic document b L 1 e b e e L 4 L 5 L 2 b b e b L 3 Line Graph e 4, b 1 L 1 b 1, e 4 L 5 b 2, e 1, b 2 L 3 Specialized Line Graph b: beginning of a Line α: Direct angle between 2 Lines e: end of a Line Eureka Meeting, L 3 i Laboratory, La Rochelle University Tuesday 20 th April 2006 Image Processing and Interpretation Group University of Nottingham

Region Graph Building Our approach (2/3) [Clementini’ 93] a b neighboring a b adjacency a b b a overlap strict and tangential inclusion [Xmin < Xmin & Xmax > Xmax] & [Ymin < Ymin & Ymax > Ymax] Ymax R 2 Ymax Ymin R 1 Ymin Xmin Eureka Meeting, L 3 i Laboratory, La Rochelle University Tuesday 20 th April 2006 Xmax Image Processing and Interpretation Group University of Nottingham

Region Graph Building Our approach (3/3) Eureka Meeting, L 3 i Laboratory, La Rochelle University Tuesday 20 th April 2006 Image Processing and Interpretation Group University of Nottingham

Conclusion and Perspectives • Conclusion • First system dealing with unformating problems • First system allowing to build region graph from large sized documents and from large sized databases • Perspectives • Extend to curves and arcs processing • Extend built graphs with neighboring relations • Reduce the unformat complexity step using a zone sorting algorithm • Use it for retrieval and indexing (not only graph building) Eureka Meeting, L 3 i Laboratory, La Rochelle University Tuesday 20 th April 2006 Image Processing and Interpretation Group University of Nottingham